Structure of the Calmodulin αII-Spectrin Complex Provides Insight into the Regulation of Cell Plasticity*

αII-spectrin is a major cortical cytoskeletal protein contributing to membrane organization and integrity. The Ca2+-activated binding of calmodulin to an unstructured insert in the 11th repeat unit of αII-spectrin enhances the susceptibility of spectrin to calpain cleavage but abolishes its sensitivity to several caspases and to at least one bacterially derived pathologic protease. Other regulatory inputs including phosphorylation by c-Src also modulate the proteolytic susceptibility of αII-spectrin. These pathways, acting through spectrin, appear to control membrane plasticity and integrity in several cell types. To provide a structural basis for understanding these crucial biological events, we have solved the crystal structure of a complex between bovine calmodulin and the calmodulin-binding domain of human αII-spectrin (Protein Data Bank ID code 2FOT). The structure revealed that the entire calmodulin-spectrin-binding interface is hydrophobic in nature. The spectrin domain is also unique in folding into an amphiphilic helix once positioned within the calmodulin-binding groove. The structure of this complex provides insight into the mechanisms by which calmodulin, calpain, caspase, and tyrosine phosphorylation act on spectrin to regulate essential cellular processes.

Spectrin ␣␤ heterodimers are organized into filamentous structures composed of two related subunits. The ␣ subunit contains 21 triple-helical homologous repeats that are further organized into five protease-sensitive domains (1). The 11th repeat is the longest and contains a unique insert that is implicated to play a critical role in spectrin integrity. The spectrinbased cytoskeleton represents a scaffold attached to both the plasma membrane and the intracellular organelles by interactions involving both integral and peripheral membrane proteins as well as by direct lipid interactions (2)(3)(4). Spectrin integrity is critical for establishing organized membrane subdomains and their maintenance, supporting integral membrane protein durability (5,6). Deletions in spectrin or dominant negative blockers of spectrin disrupt membrane-protein distributions and may lead to the total loss of critical signaling or transport proteins from the cell surface (7)(8)(9)(10). Controlled spectrin proteolysis has been detected during both apoptotic and necrotic cell death processes (11)(12)(13), upon cell shape change (14), during cell differentiation (15)(16)(17), during lens development (18), and coincident with synaptic remodeling and dendrite outgrowth (19,20). It has been observed that during the early stages of apoptosis, a characteristic 150-kDa fragment of ␣IIspectrin is generated by caspases regardless of the cell type and the apoptosis stimulus (21). In neurons, calpain-catalyzed proteolysis has been linked with N-methyl-D-aspartate receptor activation and is thought to be crucial for synaptic and neuronal plasticity. Finally, the Pet toxin, a serine proteinase from enteropathogenic bacteria, catalyzes spectrin fragmentation and the consequent collapse of the cell membrane (22)(23)(24). In each case, the cellular response ensuing from such proteolysis is determined by cleavage at a specific site in the ␣II spectrin molecule.
The 11th repeat of ␣II-spectrin contains an unusual, protease-hypersensitive insert in helix C. This insert contains recognition sites for -calpain, caspases 2, 3, and 7, and the serine proteinase Pet toxin (13,23,25,26). The N-terminal side of the 11th repeat (herein termed ␣IIspec) is flanked by an SH3 domain, whereas the C-terminal portion encompasses a calmodulin-binding domain that is proposed to be unstructured in solution (27). The Ca 2ϩ -dependent calmodulin (CaM) 3 binding to human ␣II-spectrin regulates the proteolytic susceptibility of spectrin, enhancing its cleavage by -calpain, while blocking caspase 2, 3, and 7 cleavage (28 -30). Cleavage by calpain is also controlled by phosphorylation of residue Tyr 1176 by c-Src, a kinase that binds to the flanking SH3 domain in the ␣10 repeat unit (31). When phosphorylated, ␣II-spectrin becomes resistant to calpain proteolytic activity (31,32). Thus, an insert in the ␣11 repeat of the human ␣II-spectrin represents the convergence point of two crucial cellular signaling cascades.
The mechanism(s) of both the CaM-dependent regulation of spectrin proteolysis and the cross-talk between the two signaling networks at the site of ␣IIspec is not well understood. To address these questions, we have determined the three-dimen-sional structure of the complex between bovine calmodulin and the proposed calmodulin-binding domain of human ␣II-spectrin (CaM-␣IIspec) (25). The structure of this complex revealed that the CaM-␣IIspec-binding interface is composed entirely of hydrophobic residues, in striking contrast to previously determined CaM-peptide structures. CaM binding also induced a 23-residue stretch of ␣IIspec to fold into an amphiphilic ␣-helix with hydrophobic side chains facing the CaM hydrophobic groove. These numerous hydrophobic interactions are reminiscent of those seen within a protein core, suggesting that complex formation is driven by an increase in solvent entropy. A detailed comparison of our structure with those of other CaM-peptide complex structures also reveals that ␣IIspec is positioned uniquely in the CaM groove relative to previously known ligands (33)(34)(35)(36). The crystal structure of the CaM-␣IIspec complex thus provides a compelling structural basis for understanding how calmodulin regulates the susceptibility of ␣II-spectrin to proteolysis and thereby other downstream cytoskeletal signals.

EXPERIMENTAL PROCEDURES
Complex Preparation and Crystallization-Bovine calmodulin was purchased from Sigma. The human ␣II-spectrin calmodulin-binding domain containing cleavage sites for calpain, caspases, and Pet toxin was cloned into pGEX4T vector and expressed and purified as a GST fusion protein from Escherichia coli BL21(DE3) using the following procedure. An overnight culture of the CaM-binding domain of spectrin was diluted 1/10 into fresh Luria-Bertani medium containing 100 g/ml ampicillin, grown with shaking at 37°C for 1 h when it was induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside. The growth was continued for an additional 4 h at 37°C. The cells were spun down at 5,000 ϫ g/10 min. The cell pellet was resuspended in 50 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol, 0.1 mM phenylmethylsulfonyl fluoride, 1 mM benzamidine, 0.2 mg/ml lysozyme, and 1% Triton X-100 and incubated for 30 min at 4°C with gentle rocking. The culture was then diluted with an additional 10 ml of the resuspension buffer, and while on ice, sonicated with three 10-s pulses. The lysate was then centrifuged at 18,000 rpm for 20 min. at 4°C. The soluble fraction was loaded onto a glutathione agarose matrix and washed with Tris-buffered saline buffer, and GST-␣IIspec was eluted in 4 ml of 50 mM Tris-HCl, pH 8.0, 1 mM EDTA, 1 mM dithiothreitol, and 5 mM reduced glutathione. The protein was then further purified on a calmodulin-Sepharose 4B column (Amersham Biosciences). The column was equilibrated with 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM ␤-mercaptoethanol, 0.1 mM phenylmethylsulfonyl fluoride, 1 mM CaCl 2 . Prior to loading the column, CaCl 2 was added to the sample to a final concentration of 10 mM. The column was then washed with the above buffer until the baseline absorbance was restored. The GST-␣IIspec was eluted with 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM ␤-mercaptoethanol, 0.1 mM phenylmethylsulfonyl fluoride, and 5 mM EGTA. GST was subsequently removed by thrombin cleavage and removed by absorption to glutathione-Sepharose. The ␣II-spec-CaM complex was formed from an equimolar solution of calmodulin and peptide. Crystals were obtained by the sitting drop vapor diffusion method, mixing equal volumes of the protein complex and the well solution (50 mM Tris-HCl, pH 8.0, 30% polyethylene glycol 8000, 0.1 M (NH 4 ) 2 SO 4 ) in a total volume of 40 l. The crystals belong to space group P2 1 2 1 2 1 (a ϭ 44.29, b ϭ 57.73, c ϭ 69.75 Å) with one molecule per asymmetric unit and 49% solvent content.
Data Collection, Structure Determination, and Refinement-Diffraction data were collected at room temperature on a Rigaku RAXIS II detector using a rotating copper anode as an x-ray source. The data were indexed in Denzo and scaled and reduced in Scalepack (37). The initial phases were determined by molecular replacement using the structure of calmodulin (PDB accession code: 1CDM). The search model was broken into two domains (residues 4 -70 and 79 -144), omitting the linker loop containing residues 71-80. Molecular replacement calculations were performed with Phaser 1.3 (38,39), searching for the two domains of calmodulin sequentially. The top solution had a Z score of 12 and good packing with no steric clashes. Upon a cycle of simulated annealing in CNS (40), the F o Ϫ F c map contoured at 2 revealed a three-turn helix of the spectrin calmodulin-binding domain. Further refinement was done in CNS (40), whereas electron density map inspection and manual model rebuilding were done in Quanta. Ten percent of the data was randomly assigned to an R free test set for cross-validation (41). The final stages of refinement, including individual B factor refinement, were performed against all the data. The final model yielded good geometry with 89.6% of residues in the most favored region and 10.4% in additional allowed orientations according to Procheck (42). It had an R factor of 24.5% and R free of 26.9%. The crystallographic data and final model statistics are presented in Table 1.
CD Spectroscopy-Circular dichroism studies of the purified ␣II-spectrin peptide alone, of CaM alone, and of the CaM-␣IIspec complex, either with 0.1 mM CaCl 2 or in the present of 0.1 mM EDTA, were carried out in 50 mM Tris-HCl, 20 mM NaCl, pH 8.0 at room temperature. The molar circular dichroism was measured from 190 to 260 nM, and the percentages of protein secondary structure were estimated using the program K2d available at the K2d home page (43).

RESULTS AND DISCUSSION
Overall Structure-The crystal structure of the calmodulinbinding domain of ␣II-spectrin in complex with calmodulin was solved to 2.45 Å resolution and refined to a final R factor of 24.5% (R free ϭ 26.9%) ( Table 1). The final model contains 157 protein residues, 62 water molecules, and four calcium ions. The CaM-␣IIspec complex is compact and globular, like other CaM-peptide structures solved previously (see Fig. 2A) (33)(34)(35)(36)44), with two calcium-binding lobes of CaM, designated as the N-and C-lobes, wrapped around the ␣IIspec peptide. All four calcium-binding sites are similar when compared with other CaM-ligand complex structures.
To crystallize the complex of CaM with ␣IIspec, a 42-residue peptide of the proposed calmodulin-binding domain of ␣IIspectrin was prepared (Fig. 1A). This peptide contains cleavage sites for -calpain, caspases 2, 3, and 7, and the recently identified Pet toxin proteinases (23,25,28,30). A 23-residue-long stretch (residues Ala 1189 -Arg 1211 ) of the peptide was ordered and adopted an ␣-helical conformation (Fig. 2). Five N-terminal residues of CaM, loop 70 (residues 74 -80) and residue 116, were disordered and consequently could not be modeled into the CaM-␣IIspec structure. Likewise, the 18 N-terminal residues of ␣IIspec, encompassing the calpain and caspase cleavage sites, and 3 extreme C-terminal residues, were missing from the map.
CaM-␣IIspec Binding Is Mediated Entirely by Hydrophobic Interactions-Analysis of the CD spectra of the calmodulinbinding domain of ␣II-spectrin as well as earlier modeling studies (27) show that it is disordered in solution (Fig. 3). The struc-ture of the CaM-␣IIspec complex revealed that upon calmodulin binding, ␣IIspec folds into an amphiphilic helix beginning with a proline residue at position 1191 (based on the human ␣II-spectrin numbering). The combined buried surface area at the interface between the two molecules is 2,590 Å 2 , which is a half of the ␣IIspec surface area. A more detailed analysis of ␣IIspec binding revealed a striking difference between the CaM-spectrin complex and other previously solved CaM-peptide structures.
It is well established that upon binding of Ca 2ϩ , two calmodulin lobes, designated as the N-and C-lobe, undergo a structural rearrangement to expose hydrophobic pockets and allow various ligands to bind. Flexibility in helix D (hD) that connects the two CaM lobes allows an additional conformational change upon ligand binding. In such a complex, two CaM lobes wrap around the target ligand, stabilizing the closed conformation of the CaM-peptide complex. The typical CaM ligand is a helix of about 20 residues containing two hydrophobic side chains separated by 8, 10, 14, or 16 residues that provide anchor points for the ligand (46,47). The proposed CaM-binding sequence of the ␣IIspec contains 3 hydrophobic residues, whereas the rest of the sequence does not bear any similarity with other known CaM ligands (Fig. 1B). These hydrophobic residues put the ␣IIspectrin CaM-binding domain into the 1-14 class of CaM-binding peptides, or more precisely, into its 1-8-14 subclass (47). The class/subclass name designates the spacing between the critical anchor residues in the CaM ligand. In all previously described CaM-peptide complexes, the majority of interactions with the ligand are mediated through extensive hydrogen bonding networks that are supported with limited hydrophobic interactions.
However, unlike the other CaM-binding sequences that have been characterized, the interacting surface of the CaM-␣IIspec complex is composed almost entirely of hydrophobic residues clustered into three hydrophobic pockets (Fig. 4). The first CaM-binding pocket is composed of the C-lobe residues from helices F (hF) and H (hH), the extreme C terminus of CaM, and the loop connecting hF and helix G (hG  (Fig. 4, A and B). Hydrophobic interactions between a tryptophan side chain at the N terminus of a ligand and the first hydrophobic  pocket of CaM have been described previously (34,35). In those structures, the tryptophan side chain forms a hydrogen bond with the carbonyl oxygen of Met 124 of CaM. Conversely, in the CaM-␣IIspec complex, the Trp 1192 side chain adopts a different rotamer conformation and interacts with the first methionine pocket via exclusively hydrophobic interactions (Fig. 4B). The second binding pocket is composed of residues from the N-lobe helix A (hA), C-lobe helix E (hE), and the loop connecting hF and hG. The ␣IIspec residue Leu 1197 interacts with Ala 15 and Leu 18 , whereas Val 1199 interacts with Ile 85 , Ala 88 , and Met 145 . The anchor residue in the second pocket is clearly Met 1198 , which makes contacts with Val 35 , Leu 39 , and Ala 112 (Fig. 4C). The pocket expands further with Met 36 from CaM, which interacts with side chains of Thr 1201 and Val 1202 from ␣IIspec. This pocket represents the region where the hydrophobic surfaces of the two CaM lobes fuse into one continuous region.
The third pocket is composed of residues from helices B, C, and D of the N-lobe of CaM. CaM side chains Phe 16 (Fig. 4B). Other ␣IIspec residues, such as Thr 1201 , Val 1202 , and Ile 1204 , contribute with additional Van der Waals contacts. It is clear that the hydrophobic side chains of the 2 ␣IIspec residues Trp 1192 and Phe 1205 , located at the opposite ends of the helix, interact with two methionine clefts located within different CaM lobes (Fig. 4, A and B). These residues, along with Met 1198 , serve as anchor points for the ␣IIspec helix, playing a crucial role in CaM-␣IIspec recognition.
The only two ionic interactions observed in the entire complex are the following: Lys 1193 of ␣IIspec makes a hydrogen bond with Glu 14 (N-lobe; hA), whereas Asn 1206 forms hydrogen bonds with both the Gln 41 (N-lobe; hB) and the Glu 84 (C-lobe; hE) side chains. Interestingly, these ionic interactions are contiguous to the major anchor pockets both sequentially and spatially. These small hydrogen bond networks further stabilize the orientation of the ␣IIspec helix. Finally, ␣IIspec residues from Asn 1206 to the C terminus and other polar side chains do not contribute to the binding with CaM.    Further, our data suggest that the relative orientation of the lobes differ in various complexes due to adjustments in the CaM molecule upon ligand binding. The calculated r.m.s.d. values were much higher, 2.02-2.94 Å, for the CaM-smMLCK and CaM-MARCKS complexes, respectively, when 132 C␣ atoms (residues 6 -72 and 83-147) were used for superimposition. Interestingly, calmodulin adopted a more open conformation upon binding to the ␣IIspec when compared with that seen in the previously solved complexes. The N-terminal hA and the C-lobe hG are pushed apart, and a shift is visible both in the second calciumbinding site (loop 60) and in the N-lobe hD (Fig. 5A). The flexibility of loop 60 is reflected in higher B factor values (ϳ70 Å Ϫ2 ) when compared with the rest of the molecule (ϳ46.4 Å Ϫ2 ) (Fig. 5D). It is possible that CaM had to open up more to accommodate the much longer peptide used in this study. Small spatial rearrangements of both the N-lobes and the C-lobes caused the exposure of the continuous hydrophobic binding pocket. This reflects the adaptability of CaM to different ligands. Further, because of the length of the ligand, the structure of the complex between CaM and the ␣IIspec that was derived here may be a better representation of the physiological CaM-ligand complex. The superposition of CaM molecules derived from different CaMligand complexes allowed examination of the differences in the ligand positions within the CaM-binding site.
When the CaM-␣IIspec complex is viewed in the orientation shown in Fig. 2A, it becomes evident that ␣IIspec binds differently in the CaM groove than do other CaM ligands (Fig. 5, B   FIGURE 4. The CaM-␣IIspec-binding surface is composed entirely of hydrophobic residues. A, molecular surface representations of the complex of CaM bound to ␣IIspec peptide. Left, the hydrophobic residues of CaM-binding groove are in yellow, whereas the rest of the molecule is in white. The ␣IIspec helix is shown as sticks with hydrophobic residues shown in green and polar, charged and uncharged, residues shown in blue. Right, surface charge distribution representation of CaM-␣IIspec complex. The arrows point into the two CaM methionine clefts, labeled Met cleft 1 and Met cleft 2. The N-and C-terminal residues of ␣IIspec are labeled as under "Results" (produced in Grasp (62)). The view is rotated 90 o clockwise in the plane relative to Fig. 2A. B, stereo view of the major side-chain contacts between the N-(green) and C-lobes (blue) of CaM with ␣IIspec (red). The residues, shown as balls-and-stick, are labeled as under "Results," whereas both methionine clefts are marked with an asterisk (produced in Molscript (60) and rendered in Raster 3D (61)). C, the second hydrophobic pocket positioned between two methionine clefts is composed of residues from both CaM lobes. The anchoring residues from ␣IIspec, Met 1198 and Val 1199 , interact with hydrophobic residues from both lobes of CaM. This view is rotated 90 o clockwise around the horizontal axis relative to Fig. 4B. and C). The ␣IIspec is tilted upwards toward the second methionine pocket, where the side chain of Asn 1206 provides an additional hydrogen bond network stabilizing the ligand position. This tilt of the helix is present even when compared with the CaM-smMLCK complex that bears the greatest structural similarity with CaM-␣IIspec. In all previously described complexes, both the second and the third hydrophobic pockets are either empty (CaM-NOSIII, CaM-CaMKII, and CaM-MARCKS) or occupied by a small hydrophobic residue making a limited number of Van der Waals contacts with CaM (CaM-smMLCK). Conversely, in the CaM-␣II-spec complex, there are numerous hydrophobic interactions between the ligand and the last two hydrophobic binding pockets of calmodulin. These hydrophobic interactions are reminiscent of those within the hydrophobic core of proteins, suggesting that complex formation with ␣IIspec is extremely stable and driven by increases in solvent entropy. The dissociation of CaM from ␣II-spectrin thus is likely to require dramatic changes in calmodulin structure, presumably the unfolding of calmodulin due to a fall in cellular Ca ϩ2 concentration.
Proteolysis Inhibition by Steric Hindrance and Signal Transduction by Domain Folding-The ␣11 repeat contains several proteinase recognition sites (Fig. 1A). Caspases cleave at Asp 1185 , -calpain cleaves at Tyr 1176 , and Pet toxin cleaves at Met 1198 (13,23,25). Each of these cleavages activates distinct cellular responses such as cell shape change, neuronal growth elongation, and apoptosis. The structure of the complex between CaM and the ␣IIspec provides a structural basis for understanding many of these phenomena. It is apparent that calmodulin inhibits caspase-and Pet toxin-catalyzed proteolysis of ␣II-spectrin by steric hindrance, whereas the stimulation of calpain proteolytic activity is probably accomplished through direct effects on the conformation of both the substrate and the target proteinase. The caspase cleavage site, 1182 DEDT 1185 , is only 3 residues removed from Ala 1189 , suggesting that calmodulin binding inhibits the caspase-catalyzed proteolysis of ␣II-spectrin by steric hindrance. Once the calmodulin molecule binds to ␣II-spectrin, caspase cannot approach the recognition site. It is important to note that caspase cleaves the DETD sequence efficiently, suggesting that the cleavage site assumes a substrate-like conformation when ␣IIspec is not in complex with CaM. The CaM-dependent inhibition of caspase activity abolishes the degradation of ␣␤-spectrin heterodimer and retards the completion of the apoptotic cascade. In a fashion similar to caspase inhibition, calmodulin also inhibits the Pet toxin-catalyzed proteolysis of spectrin by steric hindrance. The target residue for Pet toxin is Met 1198 . In the CaM-␣IIspec complex, Met 1198 lies buried in the second hydrophobic pocket interacting with CaM residues Val 35 , Leu 39 , and Ala 112 (Fig. 4B).
The recognition site for the cysteine proteinase milli-or -calpain is located adjacent to a putative PEST domain (48) at residue Tyr 1176 , juxtaposed to helix C of the ␣11 repeat of ␣IIspectrin (Fig. 1A). Calmodulin binding to ␣II-spectrin accelerates the rate of -calpain cleavage of ␣II-spectrin, renders an adjacent ␤II-spectrin subunit susceptible to -calpain cleavage at Gln 1441 -Ser 1442 , and modulates the quaternary association state of the post-cleaved spectrin heterodimer/heterotetramer (25,29,49). The consequences of calpain proteolysis of ␣IIspectrin are thus quite distinct from the action of the caspases, inducing the reorganization of membrane subdomains and cell shape transformation rather than cell membrane fragmentation and blebbing (13,(15)(16)(17)50). Calpain cleaves spectrin significantly only in the presence of calmodulin in vitro (30,49). Based on our structure, it is likely that these events are directly coupled to the CaM-induced folding of the ␣II-spectrin insertion containing the calpain cleavage site. The folding of the insert would constrain the conformation of the loop that contains the calpain cleavage site, forcing it to adopt a substratelike conformation in which the scissile P1-P1Ј bond (51) would assume the optimal orientation for calpain binding and proteolysis. These conclusions fit well with findings that the primary sequence of the natural calpain targets evolved to be suboptimal (52). The constraints in the spectrin insert presumably might also reorient the packing of helix C of the 11th spectrin repeat unit, triggering the conformational change that unmasks the adjacent ␤-spectrin subunit to attack by calpain.
In addition to the effects on the substrate conformation, calmodulin might affect the attacking calpain molecule through a direct interaction. The classic calpains are heterodimers consisting of large (L) and small (S) chains. The L-chain is the proteinase domain, whereas the S-chain serves as a chaperone, and the two interact through their calmodulin-like domains (53). An interesting regulatory mechanism involving dissociation of the chaperone-like S-chain has been suggested (54,55), along with the proposal that calpain activity might be regulated either by binding to endogenous activators (56) or by interaction with phospholipids (57,58) in vivo. It is plausible, therefore, that CaM interacts directly with the calmodulin-like domain of the L-chain of calpain stimulating the dissociation of the regulatory S-chain. Thus, we postulate that calmodulin might stimulate calpain activity in two ways: (i) by presenting the substrate to the proteinase in the optimal conformation and (ii) by tethering and stabilizing the tertiary structure of the attacking proteinase.
As noted above, calpain cleavage is also controlled by phosphorylation of the calpain target residue Tyr 1176 (31,32). Thus, the CaM-␣IIspec complex represents a point of convergence of at least two cellular signaling cascades. The structure of the CaM-␣IIspec complex presented here is the first attempt to understand the calmodulin-dependent regulation of ␣II-spectrin proteolysis and the downstream cytoskeletal signal transduction from a structural perspective. Further structural and biochemical studies employing progressively larger multisubunit complexes will be necessary to decipher the precise mechanisms of the cross-talk between different signaling cascades that converge at the ␣ subunit of spectrin.