The Allosteric Mechanism Induced by Protein Kinase A (PKA) Phosphorylation of Dematin (Band 4.9)*

Background: Protein kinase A regulates the stability of the erythrocyte via phosphorylation of the cytoskeletal protein dematin (band 4.9). Results: We present an experimentally derived allosteric mechanism for PKA phosphorylation of dematin. Conclusion: A phosphorylation mimicking mutation in the folded headpiece causes it to bind to the natively unfolded core domain of dematin. Significance: PKA phosphorylation of dematin represents a novel conformational switch. Dematin (band 4.9) is an F-actin binding and bundling protein best known for its role within red blood cells, where it both stabilizes as well as attaches the spectrin/actin cytoskeleton to the erythrocytic membrane. Here, we investigate the structural consequences of phosphorylating serine 381, a covalent modification that turns off F-actin bundling activity. In contrast to the canonical doctrine, in which phosphorylation of an intrinsically disordered region/protein confers affinity for another domain/protein, we found the converse to be true of dematin: phosphorylation of the well folded C-terminal villin-type headpiece confers affinity for its intrinsically disordered N-terminal core domain. We employed analytical ultracentrifugation to demonstrate that dematin is monomeric, in contrast to the prevailing view that it is trimeric. Next, using a series of truncation mutants, we verified that dematin has two F-actin binding sites, one in the core domain and the other in the headpiece domain. Although the phosphorylation-mimicking mutant, S381E, was incapable of bundling microfilaments, it retains the ability to bind F-actin. We found that a phosphorylation-mimicking mutant, S381E, eliminated the ability to bundle, but not bind F-actin filaments. Lastly, we show that the S381E point mutant caused the headpiece domain to associate with the core domain, leading us to the mechanism for cAMP-dependent kinase control of dematin's F-actin bundling activity: when unphosphorylated, dematin's two F-actin binding domains move independent of one another permitting them to bind different F-actin filaments. Phosphorylation causes these two domains to associate, forming a compact structure, and sterically eliminating one of these F-actin binding sites.

Dematin, also known as Band 4.9, is a 43-46-kDa F-actin binding and bundling protein, found in several tissue types throughout the human body (1)(2)(3). It is best known for its role within the red blood cell, where it has been localized to the junctional complex (4), a macromolecular association of at least seven different proteins that surround a short actin protofilament. These junctional complexes act as both the central hub for the approximately hexagonal spectrin lattice (5,6) as well as a point of attachment between the cytoskeleton and the erythrocytic membrane. Relative to wild-type, dematin headpiece knock-out mice have fragile, spherocytic red blood cells, which results in a compensated anemia (7). This phenotype as well as subsequent analysis (7)(8)(9) suggests that dematin is involved in anchoring the actin-spectrin cytoskeleton to the underlying membrane. In humans, this interaction is mediated through the direct association of dematin with GLUT-1, an intrinsic membrane protein (9). More recently, dematin has also been shown to stabilize the association between actin and spectrin, and this activity is turned off by PKA phosphorylation (10). The stabilizing effect that dematin has on the cytoskeleton is specifically targeted by Plasmodium spp., the pathogens causing malaria, by sequestering dematin within their parasitophorous vacuole (11). Corroborating this mechanism is the observation that the efficiency of P. falciparum infection positively correlates with cAMP levels (12), which inactivates dematin via PKA. Thus, inhibition of dematin contributes to the pathogenesis of malarial infection.
In addition to its functionality within red blood cells, dematin also plays important roles within several nucleated cell types, which are described below; however, it is less clear whether these functions stem directly from its influence on the cytoskeleton or indirectly through modulating cell signaling pathways (3,13). The dematin headpiece knock-out mouse exhibits coarse, wiry hair, and an absence of eyebrows (7,8,14). In humans, an autosomal dominant disease, Marie Unna hereditary hypotrichosis, has a very similar phenotype and results from the disruption of 8p21, the genetic locus of dematin (15). Further, the absence of dematin may also represent a significant step along the progression to prostate cancer. This is supported by the dematin gene being frequently deleted in prostate cancer (16) and, when reintroduced, it reverts the overt cancerous phenotype to that of normal prostate cells (17).
The primary sequence of dematin can be divided into two regions: an N-terminal core domain and a C-terminal "villintype" headpiece domain (Fig. 1). The headpiece domain is well folded (18,19), whereas the core domain is intrinsically disordered (20). Further, we have also shown that these two domains move independently of one another in solution (20). The lack of secondary structure in the core domain is likely the result of a high proline content (10.7%) as well as a preponderance of charged residues (122/381), both of which are characteristic of intrinsically disordered proteins (21,22).
With respect to dematin's F-actin binding and bundling activity, the literature has focused almost exclusively on its headpiece domain. Several studies have shown that dematin headpiece binds F-actin (19,23,24), as is true of nearly all headpiece domains. A unique feature of dematin headpiece is that the third to last residue is a serine (S381; glycine in all other headpiece domains) and that phosphorylation of this residue reversibly inhibits bundling activity of the full-length construct (25); however, counterintuitively, it has also been shown that phosphorylation does not inhibit the F-actin binding activity of dematin headpiece in isolation (23). In this report, we study the dynamic interaction between the dematin headpiece and the natively unfolded core domain and demonstrate that this association is controlled by PKA phosphorylation and is responsible for regulating dematin's F-actin bundling activity.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification-All constructs displayed in Fig. 1 were created from the parent construct, rD-His, (pGEX-2T vector, described in ref. 20), using QuikChange site directed mutagenesis. These constructs possess a cleavable N-terminal GST tag as well as a C-terminal His 6 tag, which were used to purify the proteins from the soluble supernatant. The His 6 tag (or any C-terminal tag) overlies the F-actin binding surface of the headpiece domain (23) and eliminates this activity, therefore, the C-terminal His 6 tag was not present in any construct used in experiments involving F-actin. Lastly, we used a phosphorylation-mimicking serine-to-glutamate (S381E) point mutant instead of directly phosphorylating the protein because even trace amounts of nonphosphorylated protein would give false positive results in the actin bundling assays. We have previously demonstrated that this point mutation very closely mimics the conformational changes within dematin headpiece that occur subsequent to phosphorylation (19), and in this report, we will demonstrate that it also reproduces the functionality.
Escherichia coli strain BL21 or BL21(DE3) cells (Novagen) were transformed with the appropriate plasmid and grown in LB medium to a cell density of ϳ0.6 -0.8 (A 600 ). The cells were then induced for 4 h by the addition of 0.8 mM isopropyl-1-thio-␤-D-galactoside. Isotopically enriched ( 15 N) proteins were obtained in a similar fashion except M9 minimal medium containing 15 N-ammonium chloride was used in lieu of rich medium (26).
Cell pellets were thawed in lysis buffer (10 mM phosphate, pH 7.0, 200 mM NaCl, 1 mg/ml lysozyme, 1 mM phenylmethylsulfonylfloride (PMSF), 1 mM EDTA, and either 5 mM ␤-mercaptoethanol or 0.5 mM DTT as reducing agents). These solutions were sonicated for 2 min on ice and centrifugally cleared. Supernatants were applied to glutathione-Sepharose 4B (Amersham Biosciences). The protein bound to the column was washed extensively with ATP buffer (10 mM phosphate, pH 7.0, 200 mM NaCl, 5 mM ATP, 5 mM MgCl 2 , 1 mM sodium azide, 0.5 mM DTT) at room temperature to remove E. coli DnaK, which tightly associates with the intrinsically disordered core domain. The column was then transferred to 4°C and equilibrated with cleavage buffer (10 mM phosphate, pH 7.0, 200 mM NaCl, 0.5 mM DTT, 1 mM sodium azide). GST fusion proteins were obtained at this point by the addition of glutathione elution buffer (50 mM Tris, pH 8.0, 5 mM reduced glutathione, 0.5 mM DTT). Constructs with the GST tag removed were collected directly off the column by on-column cleavage.
For tD and tD-S381E, the protein bound to the column was incubated with 50 units of thrombin (Amersham Biosciences) for 0.5 h at room temperature, after which, the elution was passed through p-aminobenzamidine-agarose (Sigma) to remove the protease. For rD and rD-Core constructs, 80 units of PreScission protease (Amersham Biosciences) was added, and the mixture was incubated for 4 h at 4°C prior to elution. For those constructs, which contain a C-terminal 6x-His tag, the elution in 10 mM phosphate, pH 8.0, 300 mM NaCl, 10 mM imidazole, 0.5 mM DTT was slowly passed over pre-equilibrated nickel-nitrilotriacetic acid-agarose (Qiagen), and the bound proteins were eluted with a stepwise increase of imidazole (10 -250 mM).
Actin was obtained from chicken pectoral muscle and purified from acetone powder using standard procedures (27). Prior to polymerization, the concentration of actin was determined by absorbance at 290 nm using an extinction coefficient of 26,640 M Ϫ1 ⅐cm Ϫ1 .
Analytical Ultracentrifugation-For sedimentation velocity analysis, solutions of dematin (30 M) were added to doublesector epon-filled centerpieces and analyzed with an XL-I analytical ultracentrifuge at 20°C (Beckman Coulter, Fullerton, CA). Radial absorbance scans (280 nm) were collected at a rotor speed of 40,000 rpm with radial increments of 0.003 cm in continuous scanning mode. The sedimenting boundaries were fitted to a model describing the sedimentation of a distribution of sedimentation coefficients, which were converted to mass units, c(M), using the program SEDFIT (28). Data were fitted using a regularization parameter of p ϭ 0.95, floating frictional ratios and with 200 sedimentation coefficient increments in the range of 0.5-10 S. Partial specific volumes were calculated based on amino acid composition, and buffer densities were calculated from composition using SEDNTERP (John Philo, Thousand Oaks, CA). For the fitting to a single sedimenting species, masses were fixed as monomers, and the best fit sedimentation coefficients were analyzed by Monte Carlo analysis (1000 simulations) to obtain confidence intervals.
For sedimentation equilibrium experiments, samples were loaded into 6-sector epon-filled centerpieces at a concentration of 30 M and centrifuged at rotor speeds of 10,000 rpm at 20°C until equilibrium (approximately 48 h), as assessed by the overlapping of scans taken 5 h apart. Radial scans were taken at a wavelength of 280 nm and increments of 0.001 cm in step mode. Samples were then centrifuged at 15,000 rpm for another 15 h until equilibrium was attained as indicated by overlapping scans taken at 5-h intervals. Data were analyzed in the software SEDPHAT and fitted globally for the two speeds to a model describing a single species (29) using the partial specific volume, buffer density and protein extinction coefficients calculated above with SEDNTERP.
Actin Sedimentation Assays-Increasing concentrations of pre-spun dematin stock solutions were incubated with F-actin for 1 h at 4°C. The solutions were then spun for 1 h at 100,000 ϫ g. The supernatant was removed and the pellets were solubilized overnight in SDS-PAGE loading buffer and electrophoresed through 12.5% SDS-polyacrylamide gels. F-actin sedimentation assays for isolated dematin headpiece constructs were performed and quantitated with reverse phase high performance liquid chromatography as previously described (24,30).
Actin Bundling Assay-F-actin was incubated with dematin constructs at a 1:1, weight ratio (0.2 mg/ml) at 4°C. The solutions were placed onto carbonized copper grids for 1 min, washed, and stained with 1% uranyl acetate. All samples were imaged with a Philips CM12 transmission electron microscope at 45,000ϫ magnification under minimal dose conditions. The film was processed and digitized on a Creo IQ Smart2 Scanner (Global Imaging) at 1270 dpi.
Molecular Modeling-The molecular surfaces of actin filaments (orange) were calculated from the atomic coordinates of actin (31) with the program Chimera (32) and exported to POV-RAY for further modeling (1 angstrom per POV-RAY unit) (33). Dematin was modeled as a blue prolate ellipsoid (core domain) tethered to a green sphere (headpiece) by a black linker region. The molecular size of each domain in dematin was computed from their molecular mass (35 and 10 kDa, respectively) using a value of 0.84 Da/Å 3 (34). Molecular models were rendered with POV-RAY (33).

RESULTS
Dematin Is a Monomeric Protein-Dematin has long been believed to be a trimeric protein; however, this statement was predicated only on a pseudo-3-fold appearance in rotary shadowed electron micrographs (35) and copper-phenanthroline catalyzed cysteine cross-linking experiments (36). Although we are able to reproduce these disulfide cross-linked multimers under oxidizing conditions as might be expected of any protein with a free thiol, we have been unable to cross-link them under reducing conditions with typical reagents like glutaraldehyde (data not shown). To gain a more precise view of the oligomeric status of dematin under the reducing conditions expected within the erythrocyte, we performed analytical ultracentrifugation of four constructs: (1) rD-His (core and the headpiece domain), (2) rD-S381E-His (core and the "phosphorylated" headpiece domain), (3) rD-Core-His (core domain), and (4) tD (truncated core domain and the headpiece), which are schematically represented in Fig. 1. First, we performed sedimentation velocity experiments and fitted the data to a continuous size distribution (c(M)) model (37) (Fig. 2), which produced fits indicating a monodisperse oligomeric state for each construct that was most consistent with a monomer (Fig. 2, A and B). To rule out possible weak oligomerization, which may not be evident by sedimentation velocity experiments, we also performed sedimentation equilibrium experiments (Fig. 2C). The equilibrium data, collected at two speeds, globally fitted best to models describing a monomer, which confirmed the velocity data for an absence of apparent oligomerization. Collectively, these data indicated that dematin was monomeric under these experimental conditions, which negates the current mechanism by which dematin bundles actin filaments (36,38) as this model assumed that dematin was trimeric.
Both Dematin Headpiece and the Core Domain Bind F-actin-It is well established that dematin headpiece exhibits significant affinity for F-actin (23,24); however, there also exists evidence that the core domain may also exhibit this activity (25,38). As we have demonstrated that dematin is monomeric under reducing conditions, it follows that each protein must possess two F-actin binding sites to cross-link actin filaments. To test this hypothesis, we assayed a number of constructs for F-actin binding activity via F-actin sedimentation analysis (Fig. 3). The headpiece-containing constructs, rD and tD, clearly bind in this 5 The abbreviation used is: HSQC, heteronuclear single quantum coherence. In the series of constructs called "rD" (PEST-replaced Dematin) the PEST residues were mutated as previously described (20). In another two constructs called "tD" (truncated Dematin), the residues prior to and including the PEST sequence were omitted from the transcript. All constructs were expressed as cleavable GST-fusion proteins, some of which also contain C-terminal His 6 tags.
qualitative assay. The core domain in isolation also binds F-actin in this assay (Fig. 3C), which is in accord with the results of Chishti et al., 1988 (25) and Azim et al., 1995 (38), but with much lower affinity than the headpiece-containing constructs.
Dematin (rD) but Not Phosphorylated Dematin (rD-S381E) Bundles Microfilaments-Because our data indicated that dematin is monomeric and has two F-actin binding sites, we hypothesized microfilament bundling activity must arise from this multivalency within a single polypeptide chain (Fig. 2). To test this hypothesis, we incubated both rD as well as several truncated dematin constructs with F-actin and assayed each for bundling activity via negative-stain electron microscopy (Fig.  4). We found that only rD, which contains both the native, unmodified headpiece domain as well as the core domain, could bundle microfilaments.
Previous studies have demonstrated that phosphorylation of serine 381 eliminates the F-actin bundling activity of dematin (25,36). We have previously shown that the phosphorylation mimicking mutant (S381E) very closely reproduces the conformational changes that occur secondary to phosphorylation (19), and, in Fig. 4D, we demonstrate that this modification also eliminates F-actin bundling activity. However, counterintuitively we found that the S381E mutation does not change the F-actin binding activity of dematin headpiece in isolation (Fig.  5A) (23,38).
Phosphorylation Results in a Compact Quaternary Structure-Because phosphorylation does not change the F-actin binding activity of dematin headpiece in isolation, we hypothesized that the loss of F-actin bundling activity could be explained by a phosphorylation induced conformational change. To this end, we determined the hydrodynamic properties of rD-His and rD-S381E-His from the data presented in Fig. 2A. We fitted these data to a model assuming a single sedimenting species of the monomeric mass (fixed), which is validated by the c(s) analysis and equilibrium data indicating each protein was entirely monomeric (Fig. 2, B and C). This analysis provided excellent fits with the generated sedimentation coefficients of s ϭ 2.361 S (95% CI: 2.354 S to 2.369 S) for the rD domain and s ϭ 2.379 S (95% CI: 2.367 S to 2.390 S) for the S381E mutant (Fig. 6A). These values correspond to frictional ratios (f/f 0 ) of 1.885 (95% CI: 1.879 to 1.891) and 1.872 (95% CI: 1.863 to 1.882) (Fig. 6B).
Because f/f 0 reports on the level of asymmetry (higher values represent greater asymmetry), these results suggest that the S381E mutant is more compact than the wild-type protein.
Phosphorylation Causes the Headpiece to Associate with the Core Domain-The 1 H-15 N HSQC NMR spectrum of dematin headpiece exhibited significant peak dispersion, which is characteristic of a well folded protein (18,19). In contrast, the core domain exhibited very little peak dispersion (Fig. 7A), consistent with it being intrinsically disordered (19). Furthermore, as we have previously reported, it is also missing ϳ180 backbone amide peaks, whose absence can be attributed to extreme broadening due to fast T 2 -relaxation arising from either the slow correlation (tumbling) time of the large core domain or intermediate exchange (20).
In Fig. 7, we compare the 1 H-15 N HSQC spectra of rD-His, rD-S381E-His, tD, and tD-S381E-His. We observe both the res-  onances corresponding to the natively unfolded core domain as well as the headpiece domain in all spectra except for that of rD-S381E-His, where the amide peaks arising from the headpiece domain are noticeably absent (Fig. 7B). The lack of signal from the headpiece domain can only be explained by an interaction with the large core domain, which will result in the headpiece assuming the fast T 2 relaxation time of the core domain and subsequent loss of its NMR signal. This phosphorylation dependent association between the core domain and the headpiece is schematically represented in Fig. 8. This mechanism is consistent with analytical ultracentrifugation data, which dem-onstrated that the S381E mutation results in a less asymmetric, more compact quaternary structure.

DISCUSSION
In this report, we investigated the role of dematin natively unfolded core domain with respect to its F-actin binding and bundling activity. Our results demonstrate that similar to other monomeric F-actin-binding proteins like fimbrin (39) or villin (40), dematin possesses two F-actin binding sites within a single polypeptide: one in the core domain and the other in the headpiece domain. Although the unphosphorylated headpiece domain is covalently tethered to the core domain by the peptide backbone, our NMR analysis demonstrates that their molecular motions are independent of one another. This freedom permits dematin headpiece to interact with a different actin filament than the core domain, forming parallel arrays of actin microfilaments called bundles (Fig. 8C).
In addition to endowing dematin with a second F-actin binding site and thus the ability to bundle microfilaments, the core domain is also essential for inhibiting this functionality subsequent to phosphorylation of the headpiece domain. The core domain is able to regulate dematin F-actin bundling activity because of the significant disparity in its affinity between the unphosphorylated and phosphorylated headpiece domain (phosphorylation begetting binding; Fig. 7A). In Fig. 8, we present a scaled model of dematin bound to F-actin. This level of modeling emphasizes the small size of dematin relative to actin and illustrates how the association between the headpiece and the core domain sterically prevents the protein from binding multiple actin filaments (Fig. 8).
Although we chose to use F-actin bundling activity as a prototypical reaction to demonstrate the dramatic effect that PKA phosphorylation has on the functionality of the dematin, the underlying conformational mechanism appears to control other cellular activities. For example, in mature red blood cells, which are devoid of F-actin bundles, dematin has recently been shown to stabilize the interaction between spectrin and F-actin, and this functionality is inhibited by PKA phosphorylation (10). The same molecular mechanism that we experimentally derived for actin bundling activity is easily extrapolated to tethering other ternary protein complexes linked by dematin, including spectrin and F-actin within the mature erythrocytic cytoskeleton.
In this study, we report two specific functions of dematin's intrinsically disordered core domain: (1) binding F-actin and (2) binding the "phosphorylated" headpiece domain. Although this may seem odd for a domain lacking a well defined tertiary structure, the ability to bind several different proteins is a common feature of intrinsically disordered proteins (reviewed in Ref. [41][42][43]. In fact, with the exception of enzymatic catalysis, which is nearly always performed by proteins with a well defined tertiary structure, several advantages are conferred by intrinsic disorder. These include the ability to bind a greater number of proteins due increased malleability as well as the fact that proteins lacking a well defined structure generally exhibit a larger effective molecular interface (41)(42)(43). Furthermore, in natively unfolded proteins, the act of binding is inseparable from the folding reaction, which permits fine-tuning and high-  specificity by balancing the enthalpic energy gained through binding with the loss of entropy due to folding (41)(42)(43). This high-specificity typically comes at the cost of low affinity; how-ever, in our system, this would be offset by the covalent attachment between the two domains that greatly increases their effective concentration for one another. This process might be an important aspect of the core domain's ability to discriminate between the dematin headpiece with an uncharged hydroxyl (Ser-381) relative to a negatively charged phosphate (that we mimicked with Glu-381) at a single position within dematin headpiece domain.
A significant difference between dematin and nearly all other phosphorylation-controlled conformational changes is that in dematin, phosphorylation of the well folded villin-type headpiece domain confers affinity for the intrinsically disordered core domain. This is the converse of the canonical reaction in which phosphorylation of an intrinsically disordered domain creates affinity for either a well structured domain or another intrinsically disordered domain (41)(42)(43).
Although we have reiterated that the N-terminal core domain of dematin is intrinsically disordered, it should not be assumed this is meant to denote that this domain is a randomly ordered linear chain of amino acids because, if this were true, we would have expected to observe all amide peaks in our NMR spectra due to their unrestrained motion (Fig. 7) (20). Instead, dematin likely exists as an ensemble of several metastable states each devoid of significant secondary structural elements but possessing some degree of order. This idea is corroborated by a multiple sequence alignment of dematin from several mammalian species, which reveals an extremely high degree of conservation (Ͼ94% identity). Furthermore, the intrinsically disordered core domain of dematin actually has another homolog, limatin (44), which exhibits ϳ36 and 48% identity with the core domain and entire sequence of dematin, respectively.
In summary, we present the molecular mechanism by which PKA phosphorylation regulates the functionality of dematin, using F-actin bundling activity as the prototypical reaction. When unphosphorylated, dematin's two F-actin binding domains move independently of one another permitting them to bind different F-actin filaments. Phosphorylation causes these two domains to associate, sterically eliminating one of these F-actin binding sites. Contrary to the canonical mechanism, in which phosphorylation of intrinsically disordered regions endows a protein with new binding affinities, we found the converse to be true of dematin: phosphorylation of the well folded villin-type headpiece domain conferred affinity for the intrinsically disordered core domain. Our results bring forth the notion that there are essential phosphorylation sites within well folded regions that will be systematically missed by in silico algorithms because they assume intrinsic disorder and therefore use only local sequence information. The presence of secondary structure in a globular protein obscures very local FIGURE 6. Computed sedimentation coefficients and frictional ratios. A, Plot of sedimentation coefficients (S) for rD-His, rD-S381E-His, rD-Core-His, and tD. B, Plot of frictional ratios (f/f 0 ) for rD-His, rD-S381E-His, rD-Core-His, and tD. Error bars in A and B indicate the 95% confidence intervals using Monte Carlo analysis.  . Molecular mechanism by which phosphorylation inhibits the F-actin bundling activity of dematin. A, wild-type (unphosphorylated) dematin primarily populates the "open state," a quaternary structure in which the molecular motion of dematin headpiece is largely independent of the core domain. B, in contrast, the phosphorylated version favors the "closed state", in which the phosphorylated headpiece is bound to the core domain. C, in the open state, dematin is able to bind two microfilaments and thus form bundles; however, subsequent to phosphorylation (D, closed state) dematin is only able to bind a single microfilament. sequence and thus to identify phosphorylation sites like that in dematin headpiece would require three-dimensional structural information.