Ezrin is a member of the ERM (ezrin–radixin–moesin) family of proteins that have been conserved through metazoan evolution. These proteins have dormant and active forms, where the latter links the actin cytoskeleton to membranes. ERM proteins have three domains: an N-terminal FERM [band Four-point-one (4.1) ERM] domain comprising three subdomains (F1, F2, and F3); a helical domain; and a C-terminal actin-binding domain. In the dormant form, FERM and C-terminal domains form a stable complex. We have determined crystal structures of the active FERM domain and the dormant FERM:C-terminal domain complex of human ezrin. We observe a bistable array of phenylalanine residues in the core of subdomain F3 that is mobile in the active form and locked in the dormant form. As subdomain F3 is pivotal in binding membrane proteins and phospholipids, these transitions may facilitate activation and signaling. Full-length ezrin forms stable monomers and dimers. We used small-angle X-ray scattering to determine the solution structures of these species. As expected, the monomer shows a globular domain with a protruding helical coiled coil. The dimer shows an elongated dumbbell structure that is twice as long as the monomer. By aligning ERM sequences spanning metazoan evolution, we show that the central helical region is conserved, preserving the heptad repeat. Using this, we have built a dimer model where each monomer forms half of an elongated antiparallel coiled coil with domain-swapped FERM:C-terminal domain complexes at each end. The model suggests that ERM dimers may bind to actin in a parallel fashion.

Introduction

The ezrin, radixin, and moesin (ERM) proteins connect cell membranes and the underlying actin cortex [1,2]. These three largely metazoan proteins are highly conserved paralogues (human ERM proteins have overall 73–81% sequence identity). ERM proteins are dynamically regulated by the Rho family of small GTPases, and thus, the membrane–cell cortex interaction can be switched on and off locally. ERM proteins organize membrane domains by binding to the cytoplasmic tails of integral membrane proteins [1,2], for example, in the immune synapse and in vesicular trafficking, processing, and phagocytosis. The ERM protein ezrin is directly involved in tumor metastasis, particularly in the pediatric cancers including rhabdomyosarcoma and osteosarcoma [3].

Each ERM protein contains three distinct domains [4]. The first is the highly conserved approximately 300-residue FERM domain [band Four-point-one (4.1), Ezrin, Radixin, Moesin] at the N-terminus, which has the highest (∼86%) sequence identity among the three human ERM proteins. The second is the intermediate or α-helical domain, which includes a proline-rich linker, where the latter is absent in moesin. The third domain is the C-terminal domain (sometimes referred to as the C-ERMAD) that shares approximately 71% sequence identity among human ERM proteins. The C-terminal domain contains a conserved phosphorylation site associated with activation (Thr567 in human ezrin) and a highly conserved F-actin-binding sequence in the last 30 residues at its C-terminus [5].

The ERM proteins have at least two physiologically relevant states: the dormant or autoinhibited state and the active state [6]. In the dormant state, the molecules are biologically inert, as the functional binding sites on both the FERM domain and the C-terminal domain are masked by intramolecular interactions between these domains. In the active state, these interactions are disrupted, inducing the separation of the FERM and C-terminal domains. Subsequently, the FERM domain binds to the plasma membrane, whereas the C-terminal domain binds to F-actin. In the active state, the FERM domain is able to bind directly or indirectly [for example, through an adaptor protein ERM-binding-phosphoprotein 50 (EBP50)/Na+/H+ exchanger regulatory factor (NHERF)] to the cytoplasmic extensions of membrane proteins. It has been suggested that the last two residues of ezrin C-terminus are critical for interaction with both the FERM domain and the F-actin [6].

Studies have shown that ERM proteins are phosphorylated by Rho-kinase, protein kinase Cθ, and protein kinase Cα. It has been proposed that the activation of ERM proteins in vivo occurs via a two-step mechanism [7], involving ERM proteins binding to phosphatidylinositol 4,5-biphosphate (PI(4,5)P2) and phosphorylation of the conserved threonine residue in the C-terminal domain. It has been shown that the FERM domain of ezrin interacts with liposomes containing PI(4,5)P2 [8]. Subsequent studies have shown that both phosphorylation and PI(4,5)P2 binding regulate ERM activation and thus binding to F-actin [9,10].

The first crystal structure of an ERM protein is that of moesin [11], comprising a complex between the FERM and the C-terminal domains. Subsequent crystal structures include FERM domains of ezrin [12], radixin [13], and moesin [14]; a complex between radixin and inositol-(1,4,5)-triphosphate, a mimetic of the head group of PI(4,5)P2 [13]; several complexes between the radixin FERM domain and peptides mimicking the cytoplasmic tails of membrane proteins [1519]; and complexes of radixin with the adaptor proteins NHERF-1 and NHERF-2 [20]. The only full-length crystal structure of an ERM protein is that of the monomeric dormant state of Sfmoesin from the Fall armyworm (moth) Spodoptera frugiperda [21].

Ezrin exists in various states within cells including monomers, dimers, oligomers, and heterodimers with other ERM proteins. The biological significance and function of these species is unknown. Gel filtration of Triton X-100 solubilized microvilli extracts from human JEG-3 carcinoma cells shows that ezrin is present as both a monomer and a larger homomeric species (possibly dimer or trimer) [22]. Two cytosolic species of ezrin have been purified from human placenta and have been characterized as a monomer and dimer that do not interconvert after isolation [23]. Solubilization of microvilli from human placenta show that ezrin occurs mainly as dimers and higher-order oligomers [24]. These authors also show that stimulation of human A431 carcinoma cells with EGF induces the rapid formation of ezrin oligomers in vivo, which is correlated to tyrosine phosphorylation of ezrin [24]. In the LLC-PK1 kidney epithelial cell line, endogenous ezrin is partitioned into cytosolic and membrane-associated fractions where homo-oligomers (mainly dimers) make up 15 and 28%, respectively, of the total ezrin in each fraction [25]. In contrast, they find that endogenous (threonine) phosphorylated ERM proteins are largely monomeric. Expression of wild-type and Thr567 mutant ezrins in these cell lines supports the observation that activation of ezrin by phosphorylating Thr567 produces mainly monomeric ezrin [25]. Microinjection of bacterially expressed human ezrin into live gastric HGT-1 cells shows that it co-immunoprecipitates with endogenous ERM, presumably in the form of homodimer and heterodimer [26]. Stable complexes between ezrin and moesin, presumed to be heterodimers, have been immunoprecipitated from human epidermoid carcinoma A431 cells [27]. Similarly, the ERM paralog, merlin has been co-immunoprecipitated with ezrin from human U251 glioma cells, presumably as a heterodimer [28]. The above observations show that ezrin dimers exist in mammalian cells; however, the function of the dimer is currently unknown.

In the present paper, we present the crystal structures of the active FERM domain and the inactive FERM:C-terminal domain complex of human ezrin. Comparison of these two high-resolution structures shows changes in structure and mobility that may be important in ezrin activation. In particular, the hydrophobic core of subdomain F3 in the active FERM domain structure shows a bistable structure where an array of conserved phenylalanine residues can adopt two distinct arrangements. The formation of the dormant FERM:C-terminal domain complex stabilizes this structure by selecting one arrangement for the core phenylalanine residues. We also report solution small-angle X-ray scattering (SAXS) structures of full-length ezrin monomer and dimer. The monomer structure is consistent with the full-length Sfmoesin crystal structure where the central α-helical domain protrudes from the FERM domain as an elongated α-helical coiled coil. The SAXS analysis shows that the dimer is an extremely elongated dumbbell structure. This structure is modeled by domain swapping, where the central α-helical domains from each protomer form a continuous antiparallel coiled coil linking the two domain-swapped FERM:C-terminal domain complexes. The domain-swapped nature of the dimer explains its stability and slow equilibration between monomer and dimer. Sequence analysis shows that the heptad repeat for the entire region of the central α-helical domain spanned by α-helices αB and αC, including the linker (helices are labeled as per the Sfmoesin crystal structure), is conserved throughout metazoan evolution. Thus, evolution has preserved the ability of ERM proteins to form the continuous, elongated coiled coil observed in the dimer model.

Materials and methods

Cloning

A plasmid encoding the FERM domain of human ezrin (residues 1–296 and untagged 35 kDa) was a gift from Mark Berryman (Ohio University, USA). Full-length human ezrin (residues 1–586, 69.5 kDa) was purchased from GeneArt AG, Germany, and subcloned into pET-15b vector (Novagen) for expression with an N-terminal His8-tag that can be removed by tobacco etch virus (TEV) cleavage.

Protein expression

The FERM domain of ezrin was expressed in Escherichia coli strain M15(pREP4) (Qiagen) grown in lysogeny broth (LB) media containing 100 μg/ml ampicillin and 25 μg/ml kanamycin at 37°C. Expression was induced at mid-log growth phase with 0.1 mM isopropyl β-d-thiogalactopyranoside (IPTG) at 25°C for 9–12 h.

Full-length ezrin was expressed in E. coli BL21(DE3) (Invitrogen) grown in LB media containing 100 μg/ml ampicillin at 37°C. Expression was induced at mid-log growth phase with 0.5 mM IPTG at 37°C for 3 h, producing a mixture of ezrin monomer and dimer.

Protein purification

Ezrin FERM domain

The FERM domain was purified using a modification of a previously published method [29]. The cell pellet was resuspended in lysis buffer (180 mM KH2PO4 pH 7.0) containing protease inhibitors (complete, EDTA-free protease inhibitor, Roche). Following lysis, polyethyleneimine (PEI) was added to 0.15% (v/v) after lysis to precipitate nucleic acids. The lysate was centrifuged at 30 000 g and the supernatant applied to a pre-equilibrated (with lysis buffer) hydroxyapatite column at 4°C. The protein was eluted with a gradient of 180 –900 mM KH2PO4 pH 7.0. Residual nucleic acid was removed using 0.15% (v/v) PEI and the protein was then subjected to size-exclusion chromatography (SEC) on a Superdex75 column (GE Healthcare) at 4°C in 20 mM Tris pH 8.0, 300 mM NaCl. After SEC, fractions containing the FERM domain were concentrated to approximately 12 mg/ml and frozen using liquid nitrogen.

Full-length ezrin

The cell pellet was resuspended in 20 mM Tris pH 8.0, 500 mM NaCl, 10 mM imidazole, and 0.5 mM DTT containing protease inhibitors. After lysis, PEI was added to 0.15% (v/v), and the lysate centrifuged at 30 000 g. Full-length ezrin was purified at room temperature. The clarified lysate was incubated with Ni-NTA resin (Qiagen) for 1 h. The resin was washed with 20 mM Tris pH 8.0, 500 mM NaCl, 20 mM imidazole, and 0.5 mM DTT, then protein was eluted with 20 mM Tris pH 8.0, 500 mM NaCl, 500 mM imidazole, and 0.5 mM DTT. Eluted protein was dialyzed against 20 mM Tris pH 8.0, 500 mM NaCl, 10 mM imidazole, and 1 mM DTT with a single change of dialysis solution. The His8-tag was removed by incubation with His-tagged TEV protease for 16 h. The incubated material was then passed through Ni-NTA resin. The tag-free ezrin was collected in the flow through and dialyzed against 20 mM Tris pH 8.5, 50 mM NaCl. Ezrin was applied to a Q-Sepharose column (GE Healthcare) and eluted with 20 mM Tris pH 8.5 and 145–290 mM NaCl gradient. The protein was further purified using a Superdex200 26/60 SEC column (GE Healthcare) in 20 mM HEPES pH 7.0, 100 mM KCl, and 1 mM DTT. This resulted in baseline separation of monomeric and dimeric ezrin (Supplementary Figure S1A). Purified ezrin was concentrated and frozen in liquid nitrogen.

Multiangle light scattering

Molecular masses were estimated by SEC coupled with in-line multiangle light scattering (SEC-MALS). SEC was via a Superose 12 10/300 GL column (GE Healthcare) in 20 mM HEPES pH 7.0, 100 mM KCl, and 1 mM DTT. The eluate was monitored by a miniDAWN TREOS light scattering instrument, consisting of a laser source at 690 nm and three discrete photodetectors (45°, 90°, and 135°) coupled to an Optilab DSP differential refractometer (Wyatt Technology Corporation Pty Ltd). Data were collected and analyzed using ASTRA software, and molecular weight was calculated using a refractive index increment (dn/dc) value for protein of 0.19 ml/g.

Circular dichroism

Circular dichroism (CD) experiments were performed using the Jasco J-810 CD Spectropolarimeter fitted with Peltier temperature controller. Protein was diluted in 20 mM phosphate buffer pH 7.0 to a concentration of approximately 0.2 mg/ml. Spectra were collected in the far UV (190–260 nm) with a scanning rate of 50 nm/min. The scans were repeated three times and then averaged. CD responses were converted to mean residue ellipticity [θ] (deg cm2 dmole−1).

Thermal denaturation experiments were performed at 208 nm with a heating rate of 1°C/min between 20 and 80°C. The mean residue ellipticity [θ] showed a linear dependence on temperature for both native/folded (N) and unfolded (U) states. Thus, the data were fit to two linear functions of temperature for N and U states with a sigmoidal function representing the unfolding transition:

 
formula

where TAM is the apparent melting temperature and b determines the steepness of the melting transition. The linear dependence of [θ] for N state is described by the slope m and the intercept C, whereas the linear dependence of [θ] for U state is described by the slope m+ Δm and the intercept C+ ΔC. The thermal unfolding of all proteins tested was irreversible.

Crystallization and X-ray diffraction data collection

Crystals of the ezrin FERM domain were grown at room temperature using hanging-drop, vapor-diffusion by mixing 2 μl of protein (at ∼12 mg/ml) and 2 μl of reservoir solution containing 0.2 M ammonium sulfate, 0.1 M HEPES pH 7.5, 10% isopropanol and 16–18% (w/v) PEG4000. Crystals were transferred to a cryoprotectant solution consisting of reservoir solution supplemented with 15% PEG400 and flash-frozen in liquid nitrogen.

Crystals of the ezrin FERM:C-terminal domain complex resulted from attempts to crystallize monomeric and dimeric forms of full-length ezrin at room temperature using the hanging-drop, vapor-diffusion. Crystals grew after 7–12 days in drops containing 2:1 ratio of protein (dimer: ∼10–12 mg/ml, monomer: ∼15–20 mg/ml) to reservoir solution consisting of 0.1 M Bis-Tris pH 5.5 or pH 5.9, 15–18% (w/v) PEG3350. Crystals were transferred to a cryoprotectant solution consisting of reservoir solution supplemented with either 15% PEG400 or 15% mixture of 1:1 glycerol to 50% (w/v) PEG3350. Crystals were flash-frozen in liquid nitrogen.

X-ray diffraction datasets for the ezrin FERM domain were collected at 100 K on an ADSC Quantum 210r detector installed at the Australian Synchrotron MX1 beamline. Datasets for crystals grown from full-length ezrin were collected at 100 K using an ADSC Quantum 315r detector at the Australian Synchrotron MX2 beamline. Data collection was carried out using Blu-Ice [30]. Datasets were processed with MOSFLM [31] and scaled with SCALA [32].

Crystal structure determination and refinement

Crystal structures of both the ezrin FERM domain and the ezrin FERM:C-terminal domain complex were solved by molecular replacement using Phaser [33]. The N-terminal domain of ezrin (PDB ID: 1NI2 [12]) was used as a search model for the ezrin FERM domain, whereas the Sfmoesin structure (PDB ID: 2I1J [21]) was used for crystals obtained from full-length ezrin. Model rebuilding was carried out using COOT [34]. The overall structures were refined using PHENIX [35]. Water molecules were initially built using PHENIX and then added manually during later stages of refinement. Statistics for data reduction and refinement are listed in Table 1.

Table 1
Crystallographic data and refinement statistics
PDB FERM–Ezrin
4RMA 
Ezrin-1
4RM9 
Ezrin-2
4RM8 
Data collection 
 Wavelength (Å) 0.954 0.954 0.954 
 Space group P21 C2221 P21 
 Unit cell parameters (Å) a = 48.3 a = 67.4 a = 65.0 
b = 110.6 b = 113.5 b = 111.8 
c = 66.1 c = 111.8 c = 68.4 
β = 101.6° α = β = γ = 90° β = 113.0° 
 Resolution (Å)1 1.75 (1.84–1.75) 2.0 (2.11–2.00) 1.9 (2.00–1.90) 
Rmerge (%)1 7.0 (70.1) 11.8 (80.9) 11.2 (34.2) 
 Total reflections (unique) 249 580 (66 327) 401 954 (28 759) 242 283 (67 737) 
1 10.9 (1.9) 14.1 (3.7) 7.1 (2.0) 
 Completeness (%)1 97.0 (94.6) 98.3 (97.7) 95.9 (92.0) 
 Multiplicity1 3.8 (3.6) 14.0 (14.4) 3.6 (2.0) 
 Wilson B-factor (Å2)2 21.2 (25.9, 17.9, 13.6) 24.9 (41.1, 27.5, 17.9) 24.4 (28.1, 26.7, 19.3) 
Refinement 
 Resolution range (Å) 35.97–1.75 40.24–2.00 39.40–1.90 
 Total number of atoms (protein atoms) 5437 (4942) 3233 (3086) 6776 (6198) 
 R-factor 20.0 20.2 17.4 
 R-free 24.3 25.3 21.2 
 Average B-factor (Å232.3 45.8 33.3 
 Ramachandran plot (%)3 
  Most favored 98.46% 96.2% 99.2% 
  Outliers 0.2% (Asp252-B) 0.3% (Lys63) 0.1% (Asp252-A) 
 RMS bond length (Å)4 0.013 0.003 0.005 
 RMS bond angles4 1.345° 0.736° 0.894° 
 MolProbity clash score3 4.42 6.29 3.46 
 MolProbity overall score3 1.49 1.80 1.14 
PDB FERM–Ezrin
4RMA 
Ezrin-1
4RM9 
Ezrin-2
4RM8 
Data collection 
 Wavelength (Å) 0.954 0.954 0.954 
 Space group P21 C2221 P21 
 Unit cell parameters (Å) a = 48.3 a = 67.4 a = 65.0 
b = 110.6 b = 113.5 b = 111.8 
c = 66.1 c = 111.8 c = 68.4 
β = 101.6° α = β = γ = 90° β = 113.0° 
 Resolution (Å)1 1.75 (1.84–1.75) 2.0 (2.11–2.00) 1.9 (2.00–1.90) 
Rmerge (%)1 7.0 (70.1) 11.8 (80.9) 11.2 (34.2) 
 Total reflections (unique) 249 580 (66 327) 401 954 (28 759) 242 283 (67 737) 
1 10.9 (1.9) 14.1 (3.7) 7.1 (2.0) 
 Completeness (%)1 97.0 (94.6) 98.3 (97.7) 95.9 (92.0) 
 Multiplicity1 3.8 (3.6) 14.0 (14.4) 3.6 (2.0) 
 Wilson B-factor (Å2)2 21.2 (25.9, 17.9, 13.6) 24.9 (41.1, 27.5, 17.9) 24.4 (28.1, 26.7, 19.3) 
Refinement 
 Resolution range (Å) 35.97–1.75 40.24–2.00 39.40–1.90 
 Total number of atoms (protein atoms) 5437 (4942) 3233 (3086) 6776 (6198) 
 R-factor 20.0 20.2 17.4 
 R-free 24.3 25.3 21.2 
 Average B-factor (Å232.3 45.8 33.3 
 Ramachandran plot (%)3 
  Most favored 98.46% 96.2% 99.2% 
  Outliers 0.2% (Asp252-B) 0.3% (Lys63) 0.1% (Asp252-A) 
 RMS bond length (Å)4 0.013 0.003 0.005 
 RMS bond angles4 1.345° 0.736° 0.894° 
 MolProbity clash score3 4.42 6.29 3.46 
 MolProbity overall score3 1.49 1.80 1.14 
1

Pair values correspond to overall and outer shell (in parentheses) statistics.

2

Values in parentheses correspond to maximum likelihood-based estimation of overall anisotropic Wilson B-tensor obtained using the phenix.xtriage tool from a PHENIX suite [35].

3

Values calculated using MolProbity [55].

4

From PHENIX [35].

The final model of the ezrin FERM domain contains two molecules in the asymmetric unit (molecule A — residues 2–296 and molecule B — residues 2–137, 145–146, 150–296), with Cα root mean square deviation (RMSD) of 0.39 Å. The main differences between the two molecules in the asymmetric unit are located at the flexible loop connecting α-helices α2 and α2′ in subdomain F2 (Figure 1A). In molecule B, residues 138–149 are disordered with the exception of Gly145 and Tyr146. In both molecules, the side chain of Tyr146 forms a hydrogen bond with His176; however, the orientations of Gly145 and Tyr146 differ in molecules A and B.

Crystal structure of the ezrin FERM domain and the FERM:C-terminal domain complex.

Figure 1.
Crystal structure of the ezrin FERM domain and the FERM:C-terminal domain complex.

(A) Cartoon representation of the ezrin FERM domain showing subdomains: F1 (magenta, residues 2–82), F2 (orange, residues 96–198), and F3 (green, residues 204–296), with linker regions in blue. (B) The crystal structure of the ezrin FERM:C-terminal domain complex colored as a rainbow from the N-terminus (blue) to the C-terminus (red). The orientation is similar to the FERM domain structure in (A). (C) Comparison of the three independent structures of the ezrin FERM:C-terminal domain complex (4RM8-A: green, 4RM8-B: purple, 4RM9: orange). (D) A superposition of the FERM-only structure (yellow) with the FERM:C-terminal domain complex (FERM domain: cyan, C-terminal domain: red). The panels show close-up views of the structural changes in FERM subdomain F3 (E) and F2 (F). The arrow in (E) indicates movement of the upper β-sandwich/β-sheet region in subdomain F3. Note the two close-ups are rotated relative to (D) for clarity.

Figure 1.
Crystal structure of the ezrin FERM domain and the FERM:C-terminal domain complex.

(A) Cartoon representation of the ezrin FERM domain showing subdomains: F1 (magenta, residues 2–82), F2 (orange, residues 96–198), and F3 (green, residues 204–296), with linker regions in blue. (B) The crystal structure of the ezrin FERM:C-terminal domain complex colored as a rainbow from the N-terminus (blue) to the C-terminus (red). The orientation is similar to the FERM domain structure in (A). (C) Comparison of the three independent structures of the ezrin FERM:C-terminal domain complex (4RM8-A: green, 4RM8-B: purple, 4RM9: orange). (D) A superposition of the FERM-only structure (yellow) with the FERM:C-terminal domain complex (FERM domain: cyan, C-terminal domain: red). The panels show close-up views of the structural changes in FERM subdomain F3 (E) and F2 (F). The arrow in (E) indicates movement of the upper β-sandwich/β-sheet region in subdomain F3. Note the two close-ups are rotated relative to (D) for clarity.

The structure of the FERM:C-terminal domain complex was determined in two crystal forms: space group C2221 (Ezrin-1) at 2.0 Å resolution and P21 (Ezrin-2) at 1.9 Å resolution. The former was an attempt to crystallize the full-length ezrin monomer, while the latter, the dimer. After solving the structures, it was clear that the protein had inadvertently undergone limited proteolysis during crystallization. SDS–PAGE analysis of crystals resolved two bands at approximately 37 and 10 kDa. Mass spectrometry revealed that the 37 kDa band contains the FERM domain, whereas the 10 kDa band contains the C-terminal domain, residues 510–586, with molecular weight 9336.2 Da (data not shown). Crystal structures in both space groups contain the FERM:C-terminal domain complex.

The structure of Ezrin-1 contains one molecule in the asymmetric unit (residues 1–297 and 516–586), whereas that of Ezrin-2 contains two molecules in the asymmetric unit (molecule A: residues 1–297, 516–586 and molecule B: residues 3–297, 515–586), with Cα RMSD of 0.73 Å over 365 residues. There are only small differences between Ezrin-1 and Ezrin-2 structures with RMSDs of 0.49 Å (368 residues) and 0.73 Å (365 residues) between Ezrin-1 and molecules A and B of Ezrin-2, respectively.

All co-ordinates and structure factors have been deposited with the Protein Data Bank with accession codes: 4RMA for the ezrin FERM domain, 4RM9 and 4RM8 for the FERM:C-terminal domain complexes derived from Ezrin-1 and Ezrin-2 crystals.

Sequence analysis

ERM protein sequences were obtained using the NCBI BLAST [36] server. Sequences were aligned using CLUSTAL W [37]. Phylogenetic trees were used to separate merlin protein sequences from other ERM sequences. BLAST searches restricted by taxa ensured that the final sequence set had the widest possible coverage of the ERM evolutionary tree.

Small-angle X-ray scattering

The solution structure of full-length monomeric and dimeric ezrin was determined by SEC coupled with SAXS (SEC-SAXS) using the SAXS/WAXS beamline at the Australian Synchrotron with an in-line Superose 6 PC 3.2/30 (GE Healthcare) column equilibrated with 50 mM HEPES pH 8, 200 mM Na2SO4, 50 mM K2SO4 and 5 mM β-mercaptoethanol. About 50 μl of sample (∼24 mg/ml of monomer or 10.6 mg/ml of dimer) was loaded at a flow of 0.1 ml/min, with the eluent flowing directly into the SAXS sample chamber (SEC chromatograms shown in Supplementary Figure S1B). Data were collected using a Dectris Pilatus 1M detector at wavelength of 1.13 Å. SEC-SAXS data were collected at sample-detector distances of 7 m and 650 mm.

SAXS images were taken continuously as the protein eluted from the SEC column (SEC chromatograms shown in Supplementary Figure S1B). Data were averaged over 2-s periods, reduced to one-dimensional scattering patterns, and converted to an absolute scale using the program Scatterbrain (written and provided by the Australian Synchrotron and available at http://www.synchrotron.org.au). Useful data were obtained for Q values between 5.5 × 10−3 Å−1 and 0.855 Å−1 (Q = 4π sin θ/λ, where λ is the wavelength and 2θ is the scattering angle). Data were binned in groups of five consecutive 2-s frames and averaged allowing the radius of gyration Rg and maximum chord Dmax to be estimated. Analysis of these Rg and Dmax values confirmed that the configuration of the protein was not changing over the SEC peak. For structure determination, only the 10 brightest frames were analyzed to obtain maximum contrast between protein and background scattering. Rg, Dmax, and the pair distance distribution function P(r) were calculated using Fourier methods implemented in the program GNOM [38]:

 
formula

As an independent quality check, Rg was also estimated using the Guinier approximation:

 
formula

Low and high Q data were merged using program PRIMUS [39] to produce final scattering patterns.

Ab initio methods based on a simulated annealing algorithm implemented in DAMMIF/DAMMIN [40] and GASBOR [41] were used to generate sets of models for both the ezrin monomer and dimer, using P(r) functions with input Dmax values of 172 and 335 Å, respectively. Models generated by GASBOR were favored as the program utilizes data over the entire Q range. A set of best models was selected for monomer and dimer based on χ2 with the proviso that the fit was good in the low Q region (monomer χ2 values: 0.70–1.7; dimer χ2 values: 0.7–2.2). These models were aligned by the DAMAVER program suite [42] to generate average and ‘filtered’ (volume reduced to match the expected value for molecule) shape functions. The program FoXS [43] was used to calculate SAXS profiles from atomic models. The complete experimental datasets (Qmax = 0.954 Å−1) were used to calculate χ values.

Homology modeling

Homology models for full-length ezrin monomer and dimer were generated by extending the crystal structure of the FERM:C-terminal domain complex. Sfmoesin (PDB: 2I1J and 2I1K [21]) was used as a template. A full model for human ezrin was initially generated by the I-TASSER server [44]. An ideal α-helical coiled-coil region was generated using the CCBuilder server [45]. These two models were used to supply the missing segments. Splicing was carried out manually using the program COOT [34].

Results

Structural plasticity in the crystal structure of the ezrin FERM domain

The crystal structure of the ezrin FERM domain (residues 1–296) was determined at 1.75 Å resolution (Figure 1A and Table 1) using a crystal form that is isomorphous to the previously reported 2.3 Å structure [12]. There are two copies of the FERM domain in the asymmetric unit and each molecule contains three subdomains, identified as F1 (ubiquitin fold), F2 (all α-helical acyl-CoA-binding domain fold), and F3 (phosphotyrosine binding, or pleckstrin homology domain fold), respectively, forming a clover-shaped molecule as per all ERM FERM domains [1114] (Figure 1A).

A distinguishing feature of this higher resolution structure of the ezrin FERM domain is that in one copy of the FERM domain, subdomain F3 shows two distinct conformations in the packing of an array of phenylalanine side chains in the core of the β-sandwich (Figure 2A). Four phenylalanine residues are arrayed across the core of the outer β-sheet (β-strands β5–β7) of subdomain F3: Phe250 (β5), Phe255 (β6), Phe267 (β7), and Phe269 (β7). Each of these phenylalanine residues can adopt one of the two rotamer conformations (cyan and magenta in Figure 2A). In the two conformations of the phenylalanine array in subdomain F3, each phenylalanine side chain points in the same direction like a stack of dominos. To alternate between configurations, each side chain must switch rotamer in concert with the others (Figure 2A). The refinement of the structure indicates that these conformations are equally populated in one copy of the FERM domain (molecule A, Figure 2A), whereas the other copy (molecule B, Figure 2B) shows only one conformation which is similar to the one seen in most other ERM structures (see Discussion for exceptions).

Structural plasticity in the core of subdomain F3 in the ezrin FERM domain structure.

Figure 2.
Structural plasticity in the core of subdomain F3 in the ezrin FERM domain structure.

(A–C) The array of phenylalanine residues in the core of subdomain F3 for the FERM-only structure (A: molecule A and B: molecule B both shown in green) and the FERM:C-terminal domain complex (C), where the FERM domain is shown in cyan and the C-terminal domain shown in pink. Molecule A in the FERM domain structure shows two alternative conformations of the phenylalanine array (cyan and magenta in A), whereas the other structures show a single conformation (shown in yellow in B and C). Electron density is shown as a gray mesh. (D) A cartoon representation of the crystallographic B-factors for the FERM domain structure (molecule B). Blue represents 20 Å2, whereas red represents 50 Å2. (E and F) The distribution of main chain B-factors for the outer β-sheet of subdomain F3 (β-strands β5–β7) compared with subdomain F3 and the remainder of the molecule for the ezrin FERM domain structure (E) and the FERM:C-terminal domain complex (F).

Figure 2.
Structural plasticity in the core of subdomain F3 in the ezrin FERM domain structure.

(A–C) The array of phenylalanine residues in the core of subdomain F3 for the FERM-only structure (A: molecule A and B: molecule B both shown in green) and the FERM:C-terminal domain complex (C), where the FERM domain is shown in cyan and the C-terminal domain shown in pink. Molecule A in the FERM domain structure shows two alternative conformations of the phenylalanine array (cyan and magenta in A), whereas the other structures show a single conformation (shown in yellow in B and C). Electron density is shown as a gray mesh. (D) A cartoon representation of the crystallographic B-factors for the FERM domain structure (molecule B). Blue represents 20 Å2, whereas red represents 50 Å2. (E and F) The distribution of main chain B-factors for the outer β-sheet of subdomain F3 (β-strands β5–β7) compared with subdomain F3 and the remainder of the molecule for the ezrin FERM domain structure (E) and the FERM:C-terminal domain complex (F).

The flexibility of the outer β-sheet region of subdomain F3, containing β-strands β5, β6, and β7 (Figures 1A and 2A,B), is reflected in the elevated B-factors for this region in both copies of the structure in the asymmetric unit (Figure 2D), as shown previously [12]. The main chain B-factors of β-strands β5–β7 are 20–30 Å2 higher than either the overall FERM domain or subdomain F3 (Figure 2E).

Crystal structures of the ezrin FERM:C-terminal domain complex obtained by crystallizing purified monomer and dimer samples

Crystals were obtained from both purified full-length ezrin monomer and dimer. The structures determined from these crystals were essentially identical, showing complexes between the FERM domain and the C-terminal domain with no density (or room) for the central helical domain. SDS–PAGE and mass spectrometry confirmed that limited proteolysis had occurred during crystallization (see Materials and methods).

The crystal structure of the complex between the FERM domain and the C-terminal domain was determined in two different space groups at 1.9 Å (dimer sample) and 2.0 Å (monomer sample) resolutions, respectively (Figure 1B,C, and Table 1). These structures show that the C-terminal domain (residues 516–586) is intimately wrapped around the FERM domain (residues 1–297) as seen in the previously reported structures of dormant ERMs [11,21]. The C-terminal domain is composed of four α-helices: α1C–α4C (Figure 1B).

The only differences between structures of the ezrin FERM:C-terminal domain complex are the proximity of subdomain F1 to subdomain F3 (Figure 1C). The largest difference can be seen on comparing the two molecules in the asymmetric unit of 4RM8 (Figure 1C, green and purple), where the separation between the subdomains is closer by approximately 1–2 Å in molecule B, which has lower overall B-factors.

The fact that crystals obtained from both purified monomer and dimer samples resulted in the same structure after limited proteolysis suggests that this structure is present in both forms of the protein.

Structural changes in the FERM domain on binding the C-terminal domain

There are no major changes in the structure of the FERM domain of ezrin on forming a complex with the C-terminal domain (Cα RMSDs of 0.93–1.3 Å between FERM only and FERM:C-terminal domain complexes compared with Cα RMSDs of 0.49–0.73 Å among the complexes). However, there are specific regions in the FERM domain where the binding of the C-terminal domain alters the structure (Figure 1D).

As noted from the radixin FERM domain structure [13], binding of the C-terminal domain alters the structure of the outer β-sheet (β-strands β5, β6, and β7) of subdomain F3 (Figure 1D), the region showing plasticity in the FERM-only structure. These three β-strands translate by approximately 3 Å towards subdomain F1 on binding C-terminal domain (Figure 1E). Large changes are seen in the loop connecting β-strands β6 and β7, which are directly adjacent to helix α4C in the C-terminal domain (Figure 1E).

Binding of the C-terminal domain to subdomain F3 stabilizes its plastic hydrophobic core structure. The binding of α-helix α4C positions the side chain of Phe583, so that it selects only one possible conformation for the four phenylalanine residues (Phe267, Phe269, Phe255, and Phe250) that form the phenylalanine domino array in the core of subdomain F3 (Figure 2C). This stabilization of the F3 core results in a reduction of the main chain B-factors for the outer β-sheet (β-strands β5–β7), so that they are within approximately 10 Å2 of the remainder of the structure (Figure 2F).

Ezrin FERM subdomain F2 is an all α-helical structure (helices α1, α2, α2′, α3 and α4; Figure 1). The binding of the C-terminal domain results in the rotation of helices α2′ and α3 and changes in the long loop connecting helices α2 to α2′ (Figure 1F), while helices α1, α2, and α4 make only small movements. These changes in FERM subdomain F2 accommodate the binding of helix α1C and the loop connecting it to α2C in the C-terminal domain. These changes are consistent with those seen on comparing the FERM domain structures of radixin [13] and moesin [14] with the moesin FERM:C-terminal domain complex [11].

MALS shows that bacterially expressed ezrin forms stable monomer and dimer fractions

The expression of full-length ezrin in E. coli results in two fractions that can be separated by SEC that are called monomer and dimer (Supplementary Figure S1A). To confirm that these SEC peaks actually correspond to ezrin monomer and dimer species, we performed in-line MALS with the SEC column (SEC-MALS). The estimated molecular masses of the monomer and dimer peaks were 70 800 ± 900 and 140 000 ± 2000 Da, respectively. These are consistent with the calculated molecular masses of monomer and dimer (69 470 Da for 586 residue monomer plus N-terminal glycine). Recycling either monomer or dimer fraction through the SEC column did not result in re-equilibration of the protein into monomer and dimer (data not shown). Thus, both monomer and dimer are stable under the conditions of the experiment.

Thermal stability of ezrin as measured by CD

Figure 3A shows the far UV CD spectra of the ezrin FERM domain and the monomeric and dimeric forms of full-length WT ezrin before (at 20°C) and after thermal denaturation (at 80°C). The thermal unfolding transition is not reversible. Figure 3B shows the mean residue ellipticity [θ] at 208 nm as a function of temperature for all proteins studied. They all show a clear unfolding transition. By fitting these curves with a sigmoidal function, we obtained the apparent melting temperatures, TAM, for each protein (Figure 3B). The data show that the full-length structures are significantly more stable than the isolated FERM domain (TAM of 44.8 ± 0.1°C for the FERM only, compared with 65.0 ± 1.2 and 66.2 ± 1.1°C for full-length monomer and dimer, respectively). The apparent melting temperatures of monomer and dimer species are indistinguishable. We note that prior to the sigmoidal unfolding transition, the mean residue ellipticity shows a linear dependence on temperature for both full-length monomer and dimer samples. Such a linear dependence on temperature is typical for coiled-coil proteins [46] as well as stable single α-helices [47].

CD spectroscopy and thermal denaturation of ezrin.

Figure 3.
CD spectroscopy and thermal denaturation of ezrin.

(A) CD spectra (n = 3) of proteins before (20°C showing strong features) and after (80°C showing reduced features) thermal denaturation. (B) Representative thermal denaturation curves of ezrin (FERM domain and full-length ezrin monomer and dimer) measured at 208 nm and 1°C/min heating rate, with a sigmoidal function fitted to the data.

Figure 3.
CD spectroscopy and thermal denaturation of ezrin.

(A) CD spectra (n = 3) of proteins before (20°C showing strong features) and after (80°C showing reduced features) thermal denaturation. (B) Representative thermal denaturation curves of ezrin (FERM domain and full-length ezrin monomer and dimer) measured at 208 nm and 1°C/min heating rate, with a sigmoidal function fitted to the data.

Solution structure of full-length ezrin monomer and dimer by SAXS

To gain insight into the structure of full-length ezrin, we collected SAXS data in-line with an SEC column on previously purified monomer and dimer (Figure 4 and Table 2). The scattering curves for the monomer and dimer are almost indistinguishable except at very small angles (scattering vector magnitude Q < 0.1 Å−1 corresponding to Bragg spacing of D > 63 Å). At very low angles (Q < 0.02 Å−1, D > 314 Å), the dimer scattering increases more rapidly than the monomer, as would be expected for a larger molecule (Figure 4A, inset). For Q values between 0.02 and 0.06 Å−1 (D = 314 and 105 Å, respectively), damped oscillations are observed in the difference between the monomer and dimer scattering (Figure 4A, inset). We note that a previous study failed to distinguish between the ezrin monomer and dimer via SAXS [48]. The reason for this appears to be the Q ranged used in the previous study, which had a lower limit of 0.014 Å−1, and hence, the key difference between monomer and dimer SAXS (the upsweep below 0.02 Å−1) is effectively missing in their data.

SAXS data analysis.

Figure 4.
SAXS data analysis.

(A) Logarithm of the SAXS intensity plotted as a function of the magnitude of the scattering vector, Q, for ezrin monomer (gray) overlaid on the ezrin dimer (black). The inset plots the linear difference between the scattering intensity of the dimer and monomer samples as a function of Q (the region used to calculate the linear difference is shown by the dotted box in A). (B) The scattering intensity versus Q plotted on a log–log scale. The inset shows the Guinier plots for the monomer (gray) and dimer (black). The shaded boxes indicate the data points used to calculate Rg. Note that the d values on the x-axis represent the Bragg spacings equivalent to the Q values. (C) The pair distribution plot P(r) for the monomer (gray) and dimer (black).

Figure 4.
SAXS data analysis.

(A) Logarithm of the SAXS intensity plotted as a function of the magnitude of the scattering vector, Q, for ezrin monomer (gray) overlaid on the ezrin dimer (black). The inset plots the linear difference between the scattering intensity of the dimer and monomer samples as a function of Q (the region used to calculate the linear difference is shown by the dotted box in A). (B) The scattering intensity versus Q plotted on a log–log scale. The inset shows the Guinier plots for the monomer (gray) and dimer (black). The shaded boxes indicate the data points used to calculate Rg. Note that the d values on the x-axis represent the Bragg spacings equivalent to the Q values. (C) The pair distribution plot P(r) for the monomer (gray) and dimer (black).

Table 2
SAXS analysis
 Ezrin monomer Ezrin dimer 
Data 
Q range 0.00531–0.954 Å−1 0.00531–0.954 Å−1 
D = 2π/Q 1200–6.6 Å 1200–6.6 Å 
Guinier analysis 
 Points in Guinier region (point numbers) 31 (3–33) 8 (2–9) 
Q × Rg range 0.209–0.997 0.537–0.933 
Rg (Guinier best-fit) 39.4 ± 0.3 Å 91 ± 4 Å 
P(r) analysis 
Qmax used for P(r0.5 Å−1 (points 3–375) 0.5 Å−1 (points 2–370) 
 Regularization (smoothing) parameter ‘α’5 1.0 1.0 
Dmax, best estimate, (range) 165 Å (160–175 Å) 325 Å (320–335 Å) 
Rg (P(r)), using best Dmax 41 ± 1 Å 95 ± 5 Å 
 Ezrin monomer Ezrin dimer 
Data 
Q range 0.00531–0.954 Å−1 0.00531–0.954 Å−1 
D = 2π/Q 1200–6.6 Å 1200–6.6 Å 
Guinier analysis 
 Points in Guinier region (point numbers) 31 (3–33) 8 (2–9) 
Q × Rg range 0.209–0.997 0.537–0.933 
Rg (Guinier best-fit) 39.4 ± 0.3 Å 91 ± 4 Å 
P(r) analysis 
Qmax used for P(r0.5 Å−1 (points 3–375) 0.5 Å−1 (points 2–370) 
 Regularization (smoothing) parameter ‘α’5 1.0 1.0 
Dmax, best estimate, (range) 165 Å (160–175 Å) 325 Å (320–335 Å) 
Rg (P(r)), using best Dmax 41 ± 1 Å 95 ± 5 Å 

Smoothing parameter used in GNOM [38].

Guinier (Figure 4B, inset) and P(r) analysis (Figure 4C) of the data are consistent with each other, giving the following structural parameters for the monomer: radius of gyration, Rg = 40 ± 1 Å and maximal chord, Dmax = 165 ± 5 Å; dimer: Rg = 95 ± 5 Å and Dmax = 325 ± 5 Å (Table 2). The Dmax values suggest that the ezrin dimer is twice as long as the monomer.

The P(r) plot for the monomer SAXS data shows a single peak of approximately 35 Å with an elongated tail at large pair distances (Figure 4C). In contrast, the P(r) plot for the dimer shows two peaks, one around a pair distance of 35 Å that corresponds to the monomer peak and a second, at a pair distance of approximately 210 Å (Figure 4C). The pair distribution for the dimer data is characteristic of a dumbbell-shaped protein.

Molecular models for the ezrin monomer

Ab initio models for the ezrin monomer using the SAXS data consistently resulted in comma-shaped molecules (Figure 5A). Given the similarity between these models and the crystal structure of full-length Sfmoesin [21], we used this structure combined with our crystal structure to build a complete homology model of human ezrin (Figure 5B). Modeling the central helical domain as an extended coiled coil is consistent with the experimentally determined maximum chord, Dmax, of 165 ± 5 Å for the ezrin monomer. The monomer model was docked into the envelope of the averaged (Figure 5B, gray) and filtered (Figure 5B, green) shapes generated from the family of ab initio SAXS models. Although the homology model fits the SAXS envelope, we stress that SAXS is a low-resolution technique, and thus, the envelope equally fits a family of models where the globular domain is rotated with respect to the coiled coil.

Modeling the solution structures of ezrin monomer and dimer.

Figure 5.
Modeling the solution structures of ezrin monomer and dimer.

(A) A panel of 16 ab initio reconstructions derived from the SAXS data for the ezrin monomer. These reconstructions have been aligned using the program DAMAVER [42]. (B) The homology model for the human ezrin monomer overlaid on surfaces representing the averaged (gray) and filtered (green) shapes obtained from the ab initio models as calculated by DAMAVER. (C) A panel of nine ab initio reconstructions derived from the SAXS data for the ezrin dimer. These reconstructions have been aligned using the program DAMAVER [42]. (D) The homology model for the human ezrin dimer overlaid on surfaces representing the averaged (gray) and filtered (green) shapes obtained from the ab initio models as calculated by DAMAVER. Each panel shows the models in two orientations related by a 90° rotation about the vertical axis in the plane of the image.

Figure 5.
Modeling the solution structures of ezrin monomer and dimer.

(A) A panel of 16 ab initio reconstructions derived from the SAXS data for the ezrin monomer. These reconstructions have been aligned using the program DAMAVER [42]. (B) The homology model for the human ezrin monomer overlaid on surfaces representing the averaged (gray) and filtered (green) shapes obtained from the ab initio models as calculated by DAMAVER. (C) A panel of nine ab initio reconstructions derived from the SAXS data for the ezrin dimer. These reconstructions have been aligned using the program DAMAVER [42]. (D) The homology model for the human ezrin dimer overlaid on surfaces representing the averaged (gray) and filtered (green) shapes obtained from the ab initio models as calculated by DAMAVER. Each panel shows the models in two orientations related by a 90° rotation about the vertical axis in the plane of the image.

Low-resolution/SAXS models for the ezrin dimer

Ab initio models of the ezrin dimer using SAXS data produce very elongated structures (Figure 5C). Each model has two globular regions at the extremities linked by a thinner connecting region. The panel of model structures shows that the ezrin dimer is extremely elongated with the separation between the globular structures conserved between models.

Analysis of the SAXS data shows that the maximum chord, Dmax, for the dimer is twice than that of the monomer. Additionally, the P(r) plot shows that the ezrin dimer is a dumbbell-shaped molecule. The simplest way to fulfill these two constraints is to dock two ezrin monomers, so that their coiled-coil arms align tip-to-tip so as to resemble a single coiled coil of double the monomer length. Such a model places the two FERM:C-terminal domain complexes at opposite ends of the central coiled coil producing a dumbbell-shaped molecule. The only problem with this model is that such a dimer would be highly unstable if the only dimer contact was via the very tip of the coiled coil (i.e. not a domain-swapped dimer).

Sequence analysis of the central α-helical domain

To determine the role of the central α-helical domain in dimer formation, we analyzed a wide range of ERM protein sequences. A BLAST [36] search of the nonredundant protein sequence database (NCBI) showed that ERM proteins are present in nearly all metazoa. The alignment of representative ERM protein sequences spanning the metazoan evolutionary tree shows that ERM proteins are highly conserved with many residues in the FERM and C-terminal domains invariant.

We examined the conservation of the central α-helical domain (Supplementary Figure S2). The crystal structure of the full-length ERM monomer, Sfmoesin, was used to identify the three α-helices in the α-helical domain (αA, αB, and αC; Supplementary Figure S2). The sequence of the N-terminus of the first helix, αA, is highly conserved. The length of the second α-helix, αB, appears to be conserved with the caveat that, in chordates, it appears to be longer by seven residues due to an insertion near its N-terminus. This insertion appears to have occurred during chordate evolution, as it is present in the ERM protein from the tunicate Ciona intestinalis (Supplementary Figure S2). The region between the C-terminus of α-helix αC and the C-terminal domain is poorly conserved in terms of residue identity and sequence length. Additionally, it contains low-complexity sequences.

Although the sequences of helices αB and αC are not conserved, the heptad repeat, which is a signature of coiled-coil structures [49], is conserved in these helices (Supplementary Figure S2). This can be seen in the pattern of hydrophobic residues at the ‘a’ and ‘d’ positions in the heptad repeat (highlighted in yellow in Supplementary Figure S2). More importantly, the length of the linker between α-helices αB and αC (as seen in the Sfmoesin crystal structure) is conserved [21]. The length of this linker is such that it preserves the register of the heptad repeat of the coiled coil when the sequence is extended into one long α-helix (Supplementary Figure S2). This has been noted previously, based on the alignment of the Sfmoesin sequence with the human ERM-merlin sequences [21]. By aligning ERM proteins spanning metazoa, we see that the register of the heptad repeat is conserved throughout metazoan evolution (Supplementary Figure S2).

To accommodate both the monomer and dimer forms, the linker between helices αB and αC must be able to adopt two conformations, a loop and a helical coiled coil, respectively. In this regard, we note that the residues in the heptad ‘d’ and ‘a’ positions in this loop (residues numbered 112 and 116 in Supplementary Figure S2) are not hydrophobic; however, they are observed in these core positions in two-helix antiparallel coiled coils. In the vertebrate sequences (Supplementary Figure S2), these two positions are occupied by glutamine residues (with one exception: arginine at position ‘a’ in Callorhinchus milii (elephant shark) sequence). In lower-order metazoa, the ‘d’ site (residue 112) is occupied by: Glu (6), Thr (2), Asp, or Ser; whereas the ‘a’ site (residue 116) is occupied by: Lys (6), Arg (3), or Leu. Using the CC+ relational database of coiled-coil structures [50], we find that the Swiss-Prot normalized propensities in two-helix antiparallel coiled coils for the ‘d’ site are: Gln 0.59, Glu 0.76, Thr 0.62, Asp 0.21, and Ser 0.52; while for the ‘a’ site they are: Gln 1.0, Lys 0.73, Arg 0.86, and Leu 2.66. Thus, these residues are not rare at these core positions in two-helix antiparallel coiled coils. Therefore, the link between αB and αC is consistent with a metamorphic structure that can switch between a loop and a coiled coil.

The conservation pattern of the central α-helical domain has several implications for ERM proteins. First, the preservation of the heptad repeat register through the sequence connecting α-helices αB and αC indicates that the monomer structure can be opened up to form a domain-swapped dimer where α-helices αB and αC form one, continuous α-helix that makes up half of a coiled coil. Second, the conservation of the length of α-helices αB and αC (apart from a possible seven residue insertion in chordates) indicates that the coiled-coil domain serves a biological function.

Molecular models for the ezrin dimer

Given the sequence conservation in the central α-helical domain, we used the model of the ezrin monomer to build a model for the dimer where the coiled coil is extended to form an elongated dumbbell (Figure 5D, cartoon). This domain-swapped model satisfies the requirement of doubling the maximum chord, Dmax, of the monomer. The model matches the averaged (Figure 5D, gray) and filtered (Figure 5D, green) shapes generated from the family of ab initio SAXS models. As per the monomer, SAXS is a low-resolution technique; hence, rotation of the globular domains relative to the coiled coil will not affect the match between homology model and the SAXS shapes.

The fact that the monomer and dimer SAXS profiles are nearly identical, except for very low angles, supports the domain-swapped dimer model, as it indicates that the structural features of the dimer are identical with those of the monomer, except at very long length scales. Given the crystallographic evidence for a coiled coil in the monomer [21], the high correspondence between the monomer and dimer SAXS data argues that the same coiled-coil structure has to be present in the dimer.

Evaluation of the molecular models against the SAXS data

SAXS is a low-resolution technique; hence, it is problematic to refine atomic models against SAXS data. However, SAXS data should be able to select between competing putative atomic models. To evaluate atomic models, the program FoXS was used which generates an SAXS profile by evaluating the Debye equation [43]. The program outputs a χ value as a measure of quality of fit. The full homology models for the monomer and dimer resulted in χ values of 3.92 and 4.45 (Table 3), suggesting that portions of the model are incorrect. Given this, we created a series of partial models for both monomer and dimer, starting with the crystal structure of the FERM:C-terminal domain complex, which is a true atomic structure, and increasing the completeness of the models in a stepwise fashion by adding segments starting with those with the highest confidence (Supplementary Figure S3A).

Table 3
Fitting atomic models to SAXS data
Model Crystal structure only6 Crystal structure + coiled-coil Crystal structure + coiled-coil + helix αA7 Crystal structure + coiled-coil + helix αA + β88 Full homology model 
Residues 1–297
516–586 
1–297
354–467
516–586 
1–467
516–586 
1–467
496–586 
1–586 
Completeness of model 63.1% 82.6% 92.7% 95.6% 100% 
Monomer 
χ value9 12.99 4.12 2.26 2.89 3.92 
 No. of atoms 3075 4027 4516 4659 4873 
Dimer 
χ value9 10.98 3.19 3.58 3.98 4.45 
 No. of atoms 6150 8054 9032 9318 9746 
Model Crystal structure only6 Crystal structure + coiled-coil Crystal structure + coiled-coil + helix αA7 Crystal structure + coiled-coil + helix αA + β88 Full homology model 
Residues 1–297
516–586 
1–297
354–467
516–586 
1–467
516–586 
1–467
496–586 
1–586 
Completeness of model 63.1% 82.6% 92.7% 95.6% 100% 
Monomer 
χ value9 12.99 4.12 2.26 2.89 3.92 
 No. of atoms 3075 4027 4516 4659 4873 
Dimer 
χ value9 10.98 3.19 3.58 3.98 4.45 
 No. of atoms 6150 8054 9032 9318 9746 

Crystal structure of the ezrin FERM:C-terminal domain complex (4RM8).

FERM domain continuously linked to coiled coil via helical domain helix αA plus full-helix αB.

§

C-terminal domain extended N-terminally, so that it starts with the additional β-strand β8 which is antiparallel to β-strand β5 in subdomain F3.

χ value calculated by FoXS [43].

For the monomer, the crystal structure alone gives a poor fit to the data (χ value 12.99, Table 3). Using the crystal structure of Sfmoesin as a guide, an ideal coiled coil was added to the crystal structure, reducing the χ value to 4.12. Connecting the N-terminus of the coiled coil to the FERM domain by adding helix αA and completing the N-terminus of helix αB further decreased the χ value to 2.26, which is a reasonable value for a single atomic model missing 7% of the structure [43]. The calculated SAXS profile for this model is shown in Supplementary Figure S3B along with the experimental profile. Further extension of the model reduced the quality of the fit (increased the χ value, Table 3). These segments of the model are likely to be poor as the sequence of human ezrin diverges from Sfmoesin, which is the template for building the model. They include the proline-rich segment at the C-terminus of the helical domain.

The above set of partial ezrin monomer models was used to generate dimer models via domain swapping as described above. For the dimer models, the lowest χ value was observed for a model comprising only the two copies of the crystal structure plus the coiled coil (χ value 3.19, Table 3). This is a reasonable χ value for model comprising 83% of the scattering mass [43]. Connecting the FERM domain to the coiled-coil domain increased the χ value to 3.58, in contrast with the reduction observed in the monomer model (Table 3). This suggests that either the orientations of the two globular domains of the dumbbell may be misoriented or there exists some degree of flexibility in the coiled coil, preventing a single model fitting the data. The calculated SAXS profile for the dimer model where the FERM domain is connected to the coiled coil via helix αA (equivalent to the best monomer model) is compared with the data in Supplementary Figure S3C.

Discussion

The ERM proteins form a nexus between cell membranes and the underlying cortical actin cytoskeleton. As such, they must integrate signals from several sources and transmit information. To achieve this, the ERM proteins have several distinct structural states, dormant and activated, and appear to utilize higher-order complexes including dimers. The results of our structural studies uncover potential mechanisms by which the ERM protein ezrin may transmit information between binding sites. Our SAXS-based structure for the ezrin dimer suggests a model for how ERM proteins couple membranes to F-actin so as to maintain the cortical actin cytoskeleton parallel to the membrane.

The FERM-only crystal structure represents the activated FERM domain. Our key finding is that the hydrophobic core of subdomain F3 displays a bistable structure where a row of phenylalanine residues can adopt one of two mutually exclusive configurations (Figure 2A). We are not aware of any other protein structure that displays, such a dramatic bistability, in its hydrophobic core. The core bistability is consistent with the thermal stability data (Figure 3) and the elevated B-factors (noted previously [12]; Figure 2E).

The crystal structure of the FERM:C-terminal domain complex represents the dormant form. The binding of α-helix α4C from the C-terminal domain to the edge of subdomain F3 inserts the side chain of Phe583 into the hydrophobic core of subdomain F3. This selects a single arrangement of the core phenylalanine array (Figure 2C) acting as a keystone [12] stabilizing the structure (Figure 2C,F). The binding of the adaptor proteins NHERF-1/EBP50 and NHERF-2 stabilizes subdomain F3 via the same mechanism [20].

The bistable structure of the core phenylalanine array may transmit information between functional sites on subdomain F3. This subdomain has three interaction sites: the opening between the two β-sheets adjacent to β-strand β7 and β4 where the C-terminal domain or NHERF adaptors bind; the opposite edge of the outer β-sheet (proximal to β-strand β5) where cytoplasmic domains of membrane proteins bind [1519]; and the outer surface of the subdomain F3 β-sandwich (β-strands β5, β6, and β7), which is likely to be important in membrane binding. Thus, the core phenylalanine array in subdomain F3 may be involved in integrating information among these sites.

An examination of these core phenylalanine residues in all available ERM structures shows that, in general, the phenylalanine side chains adopt the same rotamers as seen in the dormant structure. The only exceptions are radixin FERM domain bound to the cytoplasmic domain of CD44 [16] and one molecule in the radixin FERM domain dimer [51]. In both of these structures, a peptide forms an additional antiparallel β-strand that binds to β5 of subdomain F3. However, in most structures of complexes between the radixin FERM domain and peptides representing membrane protein cytoplasmic domains, the rotamer conformation of the core phenylalanine array is identical with the dormant structure [15,1719].

The precise role of the central α-helical domain in ERM proteins is enigmatic. Models for the coupling of membranes to F-actin by ERM proteins show a membrane-bound FERM domain linked to an F-actin-bound C-terminal domain via an extended central domain [2]. Studies on the isolated α-helical domain from radixin indicate that it may form a monomeric stable α-helix [52]. However, the crystal structure of Sfmoesin shows that a large portion of the α-helical domain forms a coiled coil with a conserved heptad repeat [21]. Thus, if in the activated state, the FERM and C-terminal domains were to be separated by either an extended stable α-helix or an unstructured chain, a significant hydrophobic surface would be exposed, rendering the structure unstable.

A more likely scenario is that ERM proteins maintain a coiled-coil structure in the central α-helical domain. Our CD thermal denaturation experiments on full-length ezrin show a linear dependence on temperature prior to unfolding (Figure 3B), which is typical for coiled coils [46]; however, it does not exclude single, stable α-helices [47].

The SAXS data show that the full-length, dormant ezrin monomer structure resembles that of Sfmoesin [21]. A similar result has been obtained via small-angle neutron scattering using deuterated ezrin [48]. Thus, the central α-helical domain forms a coiled coil that extends from the FERM domain which is consistent with SAXS and CD data.

The alignment of ERM protein sequences spanning metazoa shows that several features of the central α-helical domain are conserved. As noted previously, the heptad repeat is conserved, which is consistent with the coiled coil observed in the dormant state mononmer structure [21]. Additionally, the length of the region linking the two α-helices (αB and αC) in the coiled-coil structure is conserved (Supplementary Figure S2). More importantly, the length of this linker preserves the phase of the heptad repeat between the two α-helices. This is a striking feature to conserve in the central α-helical domain over the whole of metazoan evolution. It is important for understanding the nature of the ERM dimer states.

In our model for the dimer, the monomer structure is opened up at the linker region connecting α-helices αB and αC of the coiled coil. Since the linker preserves the phase of the heptad repeat, ezrin is able to form a continuous coiled coil of twice the monomer length resulting in a domain-swapped dimer. This model is consistent with previous work and it explains previous results. The first detailed characterization of the ezrin dimer using SEC and analytical ultracentrifugation ascertained that the dimer was a stable, elongated species with little exchange between species [23]. The domain-swapped nature of our dimer model explains why monomer and dimer fractions do not exchange as this would require the complete unfolding of the elongated coiled coil, which is stabilized by a large hydrophobic interface.

Several studies have revealed structures for what is likely to be the dimeric state of ERM proteins. Low-angle rotary shadowing electron microscopy revealed two forms of radixin, one predominantly in samples of wild type and the phospho-resistant T564A, which shows a globular protein, consistent with the dormant monomer structure, whereas the phosphomimetic T564E showed a large fraction of elongated dumbbell molecules [53]. Although these were originally thought to be open monomers, their properties are consistent with our dimer model.

The dimer model is also consistent with images of ezrin obtained by atomic force microscopy (AFM) [54]. These authors collected images on bacterially expressed wild type, T567A and T567D ezrin. The images show both globular structures resembling the dormant monomer and several types of dumbbell structures. The authors interpret the dumbbell structures as open and half-closed monomers; however, the images are consistent with our dimer model. We note that the AFM images also show higher-order structures.

The conservation of the linker between the two coiled-coil helices in the monomer structure so as to maintain the phase of the heptad repeat indicates that evolution has preserved the ability of ERM proteins to form elongated domain-swapped dimer structures as per our dimer model. This suggests that the ERM dimer plays a functional role in biology. The dimer structure places the two principal ERM functionalities, binding to membranes and binding to F-actin, at both ends of an elongated dumbbell. Thus, the dimer may bind along the axis of an actin filament. We speculate that such a structure may naturally link F-actin to a membrane, so that the filaments lie parallel to the membrane surface.

Abbreviations

AFM, atomic force microscopy; CD, circular dichroism; DTT, dithiothreitol; EBP50, ERM-binding-phosphoprotein 50; EGF, epidermal growth factor; ERM, ezrin–radixin–moesin; FERM, Four-point-one (4.1) ezrin–radixin–moesin; IPTG, isopropyl β-d-thiogalactopyranoside; LB, lysogeny broth; NHERF, Na+/H+ exchanger regulatory factor; PEI, polyethyleneimine; PI(4,5)P2, phosphatidylinositol 4,5-biphosphate; RMSD, root mean square deviation; SEC, size-exclusion chromatography; SEC-MALS, SEC coupled with in-line multiangle light scattering; SEC-SAXS, SEC coupled with small-angle X-ray scattering; TEV, tobacco etch virus; WT, wild type.

Author Contribution

J.M.P., K.E.W., A.P.D., and P.M.G.C. conceived and co-ordinated the study and wrote the paper. J.M.P. purified all proteins. R.B.D. performed all molecular biology tasks. J.M.P. and S.J.H. determined the crystal structures. A.P.D., A.V.S., and S.A.B. collected SAXS data. A.P.D., A.V.S., J.C.W., and P.M.G.C. analyzed SAXS data. B.C. performed MS. C.D.N. performed MALS experiments. J.M.P. performed and analyzed CD. C.G. initiated experiments on FERM. E.H.C.B. and P.M.G.C. analyzed the coiled-coil region. All authors reviewed the results and approved the final version of the manuscript.

Funding

This research was supported under the Australian National Health and Medical Research Council Project Grant scheme [APP1085718]. J.M.P. was supported by an Australian Nuclear Science and Technology Organisation (ANSTO) Postgraduate Scholarship.

Acknowledgments

This research was undertaken on the MX1, MX2, and SAXS/WAXS beamlines at the Australian Synchrotron, Victoria, Australia. SAXS experiments were carried out with the assistance of Nigel Kirby and Haydyn Mertens of the Australian Synchrotron. We thank Mark Berryman for the plasmid encoding the ezrin FERM domain, Robert Knott for assistance with preliminary SAXS measurements using the ANSTO Bruker NanoStar, Dek Woolfson for input regarding coiled coils, Joel Mackay for the usage of MALS equipment, and Sam Breit, Louise Brown, Katharine Michie, Sophia Goodchild, Till Böcking, Peter Gunning, and Thomas Huber for fruitful discussions. MS analysis was supported by access to the University of Sydney's Mass Spectrometry Core Facility.

Competing Interests

The Authors declare that there are no competing interests associated with the manuscript.

References

References
1
Neisch
,
A.L.
and
Fehon
,
R.G.
(
2011
)
Ezrin, radixin and moesin: key regulators of membrane-cortex interactions and signaling
.
Curr. Opin. Cell Biol.
23
,
377
382
doi:
2
Fehon
,
R.G.
,
McClatchey
,
A.I.
and
Bretscher
,
A.
(
2010
)
Organizing the cell cortex: the role of ERM proteins
.
Nat. Rev. Mol. Cell Biol.
11
,
276
287
doi:
3
Clucas
,
J.
and
Valderrama
,
F.
(
2014
)
ERM proteins in cancer progression
.
J. Cell Sci.
127
,
267
275
doi:
4
Bretscher
,
A.
,
Chambers
,
D.
,
Nguyen
,
R.
and
Reczek
,
D.
(
2000
)
ERM-merlin and EBP50 protein families in plasma membrane organization and function
.
Annu. Rev. Cell Dev. Biol.
16
,
113
143
doi:
5
Turunen
,
O.
,
Wahlstrom
,
T.
and
Vaheri
,
A.
(
1994
)
Ezrin has a COOH-terminal actin-binding site that is conserved in the ezrin protein family
.
J. Cell Biol.
126
,
1445
1453
doi:
6
Gary
,
R.
and
Bretscher
,
A.
(
1995
)
Ezrin self-association involves binding of an N-terminal domain to a normally masked C-terminal domain that includes the F-actin binding site
.
Mol. Biol. Cell
6
,
1061
1075
doi:
7
Fievet
,
B.T.
,
Gautreau
,
A.
,
Roy
,
C.
,
Del Maestro
,
L.
,
Mangeat
,
P.
,
Louvard
,
D.
et al. 
(
2004
)
Phosphoinositide binding and phosphorylation act sequentially in the activation mechanism of ezrin
.
J. Cell Biol.
164
,
653
659
doi:
8
Niggli
,
V.
,
Andréoli
,
C.
,
Roy
,
C.
and
Mangeat
,
P.
(
1995
)
Identification of a phosphatidylinositol-4,5-bisphosphate-binding domain in the N-terminal region of ezrin
.
FEBS Lett.
376
,
172
176
doi:
9
Nakamura
,
F.
,
Huang
,
L.
,
Pestonjamasp
,
K.
,
Luna
,
E.J.
and
Furthmayr
,
H.
(
1999
)
Regulation of F-actin binding to platelet moesin in vitro by both phosphorylation of threonine 558 and polyphosphatidylinositides
.
Mol. Biol. Cell
10
,
2669
2685
doi:
10
Yonemura
,
S.
,
Matsui
,
T.
and
Tsukita
,
S.
(
2002
)
Rho-dependent and -independent activation mechanisms of ezrin/radixin/moesin proteins: an essential role for polyphosphoinositides in vivo
.
J. Cell Sci.
115
,
2569
2580
PMID:
[PubMed]
11
Pearson
,
M.A.
,
Reczek
,
D.
,
Bretscher
,
A.
and
Karplus
,
P.A.
(
2000
)
Structure of the ERM protein moesin reveals the FERM domain fold masked by an extended actin binding tail domain
.
Cell
101
,
259
270
doi:
12
Smith
,
W.J.
,
Nassar
,
N.
,
Bretscher
,
A.
,
Cerione
,
R.A.
and
Karplus
,
P.A.
(
2003
)
Structure of the active N-terminal domain of ezrin: conformational and mobility changes identify keystone interactions
.
J. Biol. Chem.
278
,
4949
4956
doi:
13
Hamada
,
K.
,
Shimizu
,
T.
,
Matsui
,
T.
,
Tsukita
,
S.
and
Hakoshima
,
T.
(
2000
)
Structural basis of the membrane-targeting and unmasking mechanisms of the radixin FERM domain
.
EMBO J.
19
,
4449
4462
doi:
14
Edwards
,
S.D.
and
Keep
,
N.H.
(
2001
)
The 2.7 Å crystal structure of the activated FERM domain of moesin: an analysis of structural changes on activation
.
Biochemistry
40
,
7061
7068
doi:
15
Takai
,
Y.
,
Kitano
,
K.
,
Terawaki
,
S.
,
Maesaki
,
R.
and
Hakoshima
,
T.
(
2008
)
Structural basis of the cytoplasmic tail of adhesion molecule CD43 and its binding to ERM proteins
.
J. Mol. Biol.
381
,
634
644
doi:
16
Mori
,
T.
,
Kitano
,
K.
,
Terawaki
,
S.-i.
,
Maesaki
,
R.
,
Fukami
,
Y.
and
Hakoshima
,
T.
(
2008
)
Structural basis for CD44 recognition by ERM proteins
.
J. Biol. Chem.
283
,
29602
29612
doi:
17
Terawaki
,
S.-i.
,
Kitano
,
K.
and
Hakoshima
,
T.
(
2007
)
Structural basis for type II membrane protein binding by ERM proteins revealed by the radixin-neutral endopeptidase 24.11 (NEP) complex
.
J. Biol. Chem.
282
,
19854
19862
doi:
18
Takai
,
Y.
,
Kitano
,
K.
,
Terawaki
,
S.-i.
,
Maesaki
,
R.
and
Hakoshima
,
T.
(
2007
)
Structural basis of PSGL-1 binding to ERM proteins
.
Genes Cells
12
,
1329
1338
doi:
19
Hamada
,
K.
,
Shimizu
,
T.
,
Yonemura
,
S.
,
Tsukita
,
S.
and
Hakoshima
,
T.
(
2003
)
Structural basis of adhesion-molecule recognition by ERM proteins revealed by the crystal structure of the radixin-ICAM-2 complex
.
EMBO J.
22
,
502
514
doi:
20
Terawaki
,
S.-i.
,
Maesaki
,
R.
and
Hakoshima
,
T.
(
2006
)
Structural basis for NHERF recognition by ERM proteins
.
Structure
14
,
777
789
doi:
21
Li
,
Q.
,
Nance
,
M.R.
,
Kulikauskas
,
R.
,
Nyberg
,
K.
,
Fehon
,
R.
,
Karplus
,
P.A.
et al. 
(
2007
)
Self-masking in an intact ERM-merlin protein: an active role for the central α-helical domain
.
J. Mol. Biol.
365
,
1446
1459
doi:
22
Pakkanen
,
R.
and
Vaheri
,
A.
(
1989
)
Cytovillin and other microvillar proteins of human choriocarcinoma cells
.
J. Cell. Biochem.
41
,
1
12
doi:
23
Bretscher
,
A.
,
Gary
,
R.
and
Berryman
,
M.
(
1995
)
Soluble ezrin purified from placenta exists as stable monomers and elongated dimers with masked C-terminal ezrin-radixin-moesin association domains
.
Biochemistry
34
,
16830
16837
doi:
24
Berryman
,
M.
,
Gary
,
R.
and
Bretscher
,
A.
(
1995
)
Ezrin oligomers are major cytoskeletal components of placental microvilli: a proposal for their involvement in cortical morphogenesis
.
J. Cell Biol.
131
,
1231
1242
doi:
25
Gautreau
,
A.
,
Louvard
,
D.
and
Arpin
,
M.
(
2000
)
Morphogenic effects of ezrin require a phosphorylation-induced transition from oligomers to monomers at the plasma membrane
.
J. Cell Biol.
150
,
193
204
doi:
26
Andreoli
,
C.
,
Martin
,
M.
,
Le Borgne
,
R.
,
Reggio
,
H.
and
Mangeat
,
P.
(
1994
)
Ezrin has properties to self-associate at the plasma membrane
.
J. Cell Sci.
107
(
Pt 9
),
2509
2521
PMID:
[PubMed]
27
Gary
,
R.
and
Bretscher
,
A.
(
1993
)
Heterotypic and homotypic associations between ezrin and moesin, two putative membrane-cytoskeletal linking proteins
.
Proc. Natl Acad. Sci. USA
90
,
10846
10850
doi:
28
Gronholm
,
M.
,
Sainio
,
M.
,
Zhao
,
F.
,
Heiska
,
L.
,
Vaheri
,
A.
and
Carpen
,
O.
(
1999
)
Homotypic and heterotypic interaction of the neurofibromatosis 2 tumor suppressor protein merlin and the ERM protein ezrin
.
J. Cell Sci.
112
(
Pt 6
),
895
904
PMID:
[PubMed]
29
Smith
,
W.J.
and
Cerione
,
R.A.
(
2002
)
Crystallization and preliminary crystallographic analysis of the ezrin FERM domain
.
Acta Crystallogr. D Biol. Crystallogr.
58
,
1359
1361
doi:
30
McPhillips
,
T.M.
,
McPhillips
,
S.E.
,
Chiu
,
H.-J.
,
Cohen
,
A.E.
,
Deacon
,
A.M.
,
Ellis
,
P.J.
et al. 
(
2002
)
Blu-Ice and the distributed control system: software for data acquisition and instrument control at macromolecular crystallography beamlines
.
J. Synchrotron Radiat.
9
,
401
406
doi:
31
Leslie
,
A.G.W.
(
1992
)
Recent changes to the MOSFLM package for processing film and image plate data. Joint CCP4 + ESF-EAMCB News-letter of Protein Crystallography, No. 26. CCP4, CCLRC Daresbury Laboratory, Cheshire, UK
32
Collaborative Computational Project No. 4
(
1994
)
The CCP4 suite: programs for protein crystallography
.
Acta Crystallogr. D Biol. Crystallogr.
50
,
760
763
doi:
33
McCoy
,
A.J.
,
Grosse-Kunstleve
,
R.W.
,
Adams
,
P.D.
,
Winn
,
M.D.
,
Storoni
,
L.C.
and
Read
,
R.J.
(
2007
)
Phaser crystallographic software
.
J. Appl. Crystallogr.
40
,
658
674
doi:
34
Emsley
,
P.
,
Lohkamp
,
B.
,
Scott
,
W.G.
and
Cowtan
,
K.
(
2010
)
Features and development of Coot
.
Acta Crystallogr. D Biol. Crystallogr.
66
,
486
501
doi:
35
Adams
,
P.D.
,
Afonine
,
P.V.
,
Bunkóczi
,
G.
,
Chen
,
V.B.
,
Davis
,
I.W.
,
Echols
,
N.
et al. 
(
2010
)
PHENIX: a comprehensive Python-based system for macromolecular structure solution
.
Acta Crystallogr. D Biol. Crystallogr.
66
,
213
221
doi:
36
Altschul
,
S.F.
,
Madden
,
T.L.
,
Schaffer
,
A.A.
,
Zhang
,
J.
,
Zhang
,
Z.
,
Miller
,
W.
et al. 
(
1997
)
Gapped BLAST and PSI-BLAST: a new generation of protein database search programs
.
Nucleic Acids Res.
25
,
3389
3402
doi:
37
Thompson
,
J.D.
,
Higgins
,
D.G.
and
Gibson
,
T.J.
(
1994
)
CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice
.
Nucleic Acids Res.
22
,
4673
4680
doi:
38
Svergun
,
D.
(
1992
)
Determination of the regularization parameter in indirect-transform methods using perceptual criteria
.
J. Appl. Crystallogr.
25
,
495
503
doi:
39
Konarev
,
P.V.
,
Volkov
,
V.V.
,
Sokolova
,
A.V.
,
Koch
,
M.H.J.
and
Svergun
,
D.I.
(
2003
)
PRIMUS: a Windows PC-based system for small-angle scattering data analysis
.
J. Appl. Crystallogr.
36
,
1277
1282
doi:
40
Franke
,
D.
and
Svergun
,
D.I.
(
2009
)
DAMMIF, a program for rapid ab-initio shape determination in small-angle scattering
.
J. Appl. Crystallogr.
42
,
342
346
doi:
41
Svergun
,
D.I.
,
Petoukhov
,
M.V.
and
Koch
,
M.H.
(
2001
)
Determination of domain structure of proteins from X-ray solution scattering
.
Biophys. J.
80
,
2946
2953
doi:
42
Volkov
,
V.V.
and
Svergun
,
D.I.
(
2003
)
Uniqueness of ab initio shape determination in small-angle scattering
.
J. Appl. Crystallogr.
36
,
860
864
doi:
43
Schneidman-Duhovny
,
D.
,
Hammel
,
M.
,
Tainer
,
J.A.
and
Sali
,
A.
(
2016
)
FoXS, FoXSDock and MultiFoXS: Single-state and multi-state structural modeling of proteins and their complexes based on SAXS profiles
.
Nucleic Acids Res.
doi:
44
Zhang
,
Y.
(
2008
)
I-TASSER server for protein 3D structure prediction
.
BMC Bioinformatics
9
,
40
doi:
45
Wood
,
C.W.
,
Bruning
,
M.
,
Ibarra
,
A.A.
,
Bartlett
,
G.J.
,
Thomson
,
A.R.
,
Sessions
,
R.B.
et al. 
(
2014
)
CCBuilder: an interactive web-based tool for building, designing and assessing coiled-coil protein assemblies
.
Bioinformatics
30
,
3029
3035
doi:
46
Holtzer
,
M.E.
and
Holtzer
,
A.
(
1992
)
Alpha-helix to random coil transitions: interpretation of the CD in the region of linear temperature dependence
.
Biopolymers
32
,
1589
1591
doi:
47
Wang
,
E.
and
Wang
,
C.-L.
(
1996
)
(I, i + 4) Ion pairs stabilize helical peptides derived from smooth muscle caldesmon
.
Arch. Biochem. Biophys.
329
,
156
162
doi:
48
Jayasundar
,
J.J.
,
Ju
,
J.H.
,
He
,
L.
,
Liu
,
D.
,
Meilleur
,
F.
,
Zhao
,
J.
et al. 
(
2012
)
Open conformation of ezrin bound to phosphatidylinositol 4,5-bisphosphate and to F-actin revealed by neutron scattering
.
J. Biol. Chem.
287
,
37119
37133
doi:
49
Cohen
,
C.
and
Parry
,
D.A.D.
(
1986
)
Alpha-helical coiled coils — a widespread motif in proteins
.
Trends Biochem. Sci.
11
,
245
248
doi:
50
Testa
,
O.D.
,
Moutevelis
,
E.
and
Woolfson
,
D.N.
(
2009
)
CC+: a relational database of coiled-coil structures
.
Nucleic Acids Res.
37
,
D315
D322
doi:
51
Kitano
,
K.
,
Yusa
,
F.
and
Hakoshima
,
T.
(
2006
)
Structure of dimerized radixin FERM domain suggests a novel masking motif in C-terminal residues 295-304
.
Acta Crystallogr. Sect. F Struct. Biol. Cryst. Commun.
62
,
340
345
doi:
52
Hoeflich
,
K.P.
,
Tsukita
,
S.
,
Hicks
,
L.
,
Kay
,
C.M.
and
Ikura
,
M.
(
2003
)
Insights into a single rod-like helix in activated radixin required for membrane–cytoskeletal cross-linking
.
Biochemistry
42
,
11634
11641
doi:
53
Ishikawa
,
H.
,
Tamura
,
A.
,
Matsui
,
T.
,
Sasaki
,
H.
,
Hakoshima
,
T.
and
Tsukita
,
S.
(
2001
)
Structural conversion between open and closed forms of radixin: low-angle shadowing electron microscopy
.
J. Mol. Biol.
310
,
973
978
doi:
54
Liu
,
D.
,
Ge
,
L.
,
Wang
,
F.
,
Takahashi
,
H.
,
Wang
,
D.
,
Guo
,
Z.
et al. 
(
2007
)
Single-molecule detection of phosphorylation-induced plasticity changes during ezrin activation
.
FEBS Lett.
581
,
3563
3571
doi:
55
Chen
,
V.B.
,
Arendall
, III,
W.B.
,
Headd
,
J.J.
,
Keedy
,
D.A.
,
Immormino
,
R.M.
,
Kapral
,
G.J.
et al. 
(
2010
)
Molprobity: all-atom structure validation for macromolecular crystallography
.
Acta Crystallogr. D Biol. Crystallogr.
66
,
12
21
doi:

Author notes

*

Present address: MX Beamlines, Australian Synchrotron, 800 Blackburn Road, Clayton, VIC 3168, Australia.

Present address: The Victor Chang Cardiac Research Institute, Darlinghurst, NSW 2010, Australia.

Supplementary data