Competition between Coiled-Coil Structures and the Impact on Myosin-10 Bundle Selection

Abstract

Coiled-coil fusions are a useful approach to enforce dimerization in protein engineering. However, the final structures of coiled-coil fusion proteins have received relatively little attention. Here, we determine the structural outcome of adjacent parallel and antiparallel coiled coils. The targets are coiled coils that stabilize myosin-10 in single-molecule biophysical studies. We reveal the solution structure of a short, antiparallel, myosin-10 coiled-coil fused to the parallel GCN4-p1 coiled coil. Surprisingly, this structure is a continuous, antiparallel coiled coil where GCN4-p1 pairs with myosin-10 rather than itself. We also show that longer myosin-10 segments in these parallel/antiparallel fusions are dynamic and do not fold cooperatively. Our data resolve conflicting results on myosin-10 selection of actin filament bundles, demonstrating the importance of understanding coiled-coil orientation and stability.

Introduction

Myosin-10 transports several cell-surface receptors to filopodial tips in vertebrate cells (1, 2, 3). These filopodia are actin-based protrusions at the plasma membrane with a central core of a fascin-actin bundle (4, 5). In earlier work, we defined a bundle selection mechanism for myosin-10 that explains how it navigates to filopodia and functions as an intrafilopodial transporter (2). Specifically, myosin-10 is maximally processive when it straddles two filaments within the fascin-actin bundle, and walks poorly along single filaments (6, 7, 8). This bundle selectivity is specific to myosin-10, as myosin-5 walks equally well along both bundles and single filaments (6).

However, the concept of myosin-10 as a bundle selective motor has been controversial. Two other groups (9, 10, 11) found that myosin-10 is a processive transporter on both single filaments and bundles. All three groups made truncated myosin-10 heavy meromyosin (HMM) fragments using different coiled-coil lengths and design patterns (see Fig. 1 A; Fig. S1 in the Supporting Material). Specifically, all three designs include a C-terminal stabilizing coiled coil (12). Two designs use GCN4-p1, a 34-amino acid leucine-zipper derived from the GCN4 yeast transcription activator (13). The third design uses the myosin-5 coiled coil for stabilization. Such stabilization is required in low-concentration single-molecule studies. Because we identified the coiled coil as the critical element that governs bundle selection (7), we investigate here the structures of the three excised coiled-coils (M10^short-GCN4, M10^long-GCN4, and M10^long-M5CC; see Figs. 1, A and B, and S2).

Figure 1.

Prior myosin-10 work fused parallel and antiparallel coiled coils. (A) Construct map of full-length myosin-10 (top), the three full-length constructs used in prior single-molecule work (middle), and the three coiled-coil constructs used in this study (bottom, M5: myosin-5; M10, myosin-10; SAH, single α-helix (57), BD, binding-domain). Note the differences in the length of the native myosin-10 coiled coil and the type of stabilizing C-terminal coiled-coil (GCN4 vs. M10CC) between all three constructs. (B) Paircoil2 (58) scores for the WT myosin-10 and the three coiled-coil constructs. P-scores below a threshold of 0.02 indicate a predicted coiled coil. The yellow zone indicates the defined bounds of the WT coiled coil. M10^short-GCN4 and M10^long-GCN4 have a continuous heptad repeat across the boundary, whereas M10^long-M5CC (right, note the larger residue range) has a break at the boundary. (C) Illustration of the myosin-10 coiled-coil solution NMR structure determined by the Zhang group (14). The N-terminal helices (αA) are antiparallel, whereas the C-terminal helices (αB) bend over the ends. The SAH domains (57), IQ domains, and motor domains extend from the N-terminal ends. Locations of the last myosin-10 coiled-coil residue for the M10^short-GCN4, M10^long-M5CC, and M10^long-GCN4 constructs are shown. The stabilizing coiled coils immediately follow the C-terminal end of each construct. Note that the myosin-10 portion of M10^short-GCN4 terminates early in the αB helix, whereas M10^long-M5CC and M10^long-GCN4 both include the entire αB helix. (D) Five proposed structural outcomes when a myosin-10 coiled coil (blue) is fused to a C-terminal GCN4-p1/myosin-5 coiled coil (orange). To see this figure in color, go online.

The myosin-10 coiled coil is of particular interest because an NMR structure shows that it is antiparallel with a C-terminal bend (14) (see Fig. 1 C). This antiparallel orientation is currently unique in dimeric myosins. M10^short-GCN4, M10^long-M5CC, and M10^long-GCN4 HMM constructs all fuse their stabilizing coiled coil at the C-terminus of the myosin-10 coiled coil. However, all three designs had inadvertently fused a parallel coiled coil to an antiparallel coiled coil.

The final structure of fused parallel and antiparallel coiled coils is uncertain. The possibilities include: overall parallel, overall antiparallel, partially folded or frustrated states, or both parallel and antiparallel segments with intervening extended linkers (which we call the “genie bottle” configuration in reference to its shape; see Fig. 1 D). Due to the incompatibility of parallel and antiparallel coiled coils, we envision a structural competition between states, where the ultimate outcome is determined by the balance of favorable and unfavorable interactions. In this structural competition, maximal burial of hydrophobic surface area will be a key consideration.

Here, we solve the solution NMR structure of M10^short-GCN4, and show that it forms an antiparallel coiled coil. This polarity is surprising because M10^short-GCN4 contains GCN4-p1, a coiled-coil sequence that is parallel on its own. Moreover, the somewhat longer M10^long-GCN4 has conformational dynamics on the NMR timescale and does not fold with an all-or-none transition, whereas the much longer M10^long-M5CC has structural defects that lead to oligomerization as determined by small-angle x-ray scattering (SAXS). These structural differences help to explain the conflicting functional studies of the three groups.

Materials and Methods

Vector design and cloning

Myosin-10 (Myo10), myosin-5, and GCN4 DNA sequences were polymerase chain reaction amplified with homologous ends to create the M10^short-GCN4 and M10^long-M5CC constructs (6, 9) (Fig. 1). Sequences were designed with the coiled coil immediately following the N-terminal thrombin cleavage tag, with no extra C-terminal residues. Polymerase chain reactions were performed using Pfu Turbo polymerase (Agilent, Santa Clara, CA). M10^long-GCN4 was made from a gBlock (Integrated DNA Technologies, Coralville, IA) containing myosin-10 coiled-coil, and GCN4-p1 DNA fragments (11). These constructs were cloned into pET-15b (Novagen, Darmstadt, Germany) linearized by NdeI (New England Biolabs, Ipswich, MA) following the Gibson assembly (15) (for M10^short-GCN4) and SLiCE cloning (16) protocols (for M10^long-M5CC and M10^long-GCN4) with homemade cloning mixes.

Protein expression

Coiled coils were expressed in BL21(DE3) pLysS Escherichia coli cells (EMD Millipore, Darmstadt, Germany) grown in either terrific broth or M9 minimal media (for NMR studies). After overnight expression, cells were lysed in high-salt lysis buffer (50 mM HEPES, 500 mM NaCl, 10 mM imidazole, 5% v/v glycerol, 5 mM BME, pH = 7.5) using microfluidization. The lysate was clarified by centrifugation and applied to a 30 mL Ni-NTA Superflow (Clontech, Mountain View, CA) column. The column was washed with 5 CV wash buffer (lysis buffer with 20 mM imidazole), 5 CV second wash buffer (lysis buffer with 50 mM imidazole), and 6 CV elution buffer (lysis buffer with 250 mM imidazole). The fractions with protein were pooled, concentrated, and desalted into thrombin cleavage buffer (20 mM Tris-HCl and 100 mM NaCl, pH 7.5) using an Amicon Ultra Ultracel-3 (EMD Millipore). The His-tag was removed by cleavage with human α-thrombin (Haematological Technologies, Essex Junction, VT) at room temperature for 4.5 h with nutation. The thrombin reaction was quenched with phenylmethanesulfonyl fluoride, and cleaved protein was separated from uncleaved protein and the (His)₆ tag by another pass over Ni-NTA, retaining the flowthrough. Protein was pooled and concentrated using an Amicon Ultra Ultracel-3 and was further purified by gel-filtration on a Superdex 75 30/100 GL column. Buffer conditions for gel filtration and further dialysis steps are detailed below for each experiment. The gel filtration column was calibrated with a reference set of protein markers (Sigma MWFGF70, St. Louis, MO).

NMR measurements

Proteins were prepared in 100 mM potassium phosphate, 1 mM EDTA, pH 6.5. Pooled and concentrated fractions were dialyzed in gel filtration buffer plus 0.03% w/v sodium azide. Chemical shift assignments were made using standard protein NMR methodology. Preliminary HSQC-TROSY experiments were conducted on a 600 MHz Bruker spectrometer (Billerica, MA) at the Biomolecular NMR Core at the University of Chicago. Final NMR experiments were executed at the University of Minnesota NMR Center on a Bruker Ascend 850 MHz spectrometer equipped with a TCl CryoProbe, with the sample temperature held constant at 308 K. The HN(CA)CO experiment was performed on a Varian 600 MHz spectrometer equipped with a ¹³C enhanced cold HCN Z-gradient probe. HNCA, HN(CO)CA, HNCO, HN(CA)CO CBCA(CO)NH, and CBCANH experiments were used to assign the backbone atom resonances for residues 883 through 949. Likewise, HCCCONH, HBCBCGCDHD, HCCH-TOCSY, and HCCH-COSY spectra were used to assign the side-chain ¹H and ¹³C resonances for residues 884 through 949, all the M10^short-GCN4 residues. Distance restraints were obtained from 3D ¹⁵N-NOESY (200 ms mixing time) and 3D ¹³C-NOESY (200 ms mixing time) experiments. Data were processed using NMRPipe (17) and analyzed using the CCPNMR software suite (18).

Nuclear Overhauser effect (NOE) peak volumes were calibrated using the average NOE volume from geminal Hβ atoms and classified into the following distance bins: short (1.8–3.7), medium (1.8–5.0), and long (1.8–6.0). Dihedral constraints were calculated using TALOS+ software (19). NOE distance constraints were imported into XPLOR-NIH and used for NMR structure calculation (20). We generated a composite containing 10 lowest energy structures from 100 calculated structures. We used The Protein Structure Validation Suite (PSVS), version 1.5 (21) to validate structures, and VMD and VMD-Xplor to generate molecular representations (22). The SBGrid Consortium provided software binaries (23).

SAXS

Samples were purified as described previously, using 50 mM Tris-HCl and 100 mM KCl at pH 7.5 for the gel filtration step. The samples were concentrated and dialyzed into gel filtration buffer containing 5% v/v glycerol. M10^long-M5CC samples contained 5 mM 2-mercaptoethanol as a reducing agent during the purification steps and 5 mM DTT during the dialysis and data collection steps. X-ray scattering data as a function of the momentum transfer were collected on beamline 12-ID-B at the Advanced Photon Source, Argonne National Laboratory. Samples were purified as described previously and syringe filtered before being loaded into a capillary tube. A Hamilton syringe pump agitated the samples in the capillary during x-ray exposure to minimize sample damage. M10^short-GCN4 was exposed at 15 mg/mL, 7.5 mg/mL, and 3.8 mg/mL, and M10^long-M5CC was exposed at 12 mg/mL, 6 mg/mL, and 3 mg/mL to create a dilution series. SAXS data were collected continuously over a range of Q from 0.004 to 1.00 Å^-1. Buffer blanks (50 mM Tris-HCl, 100 mM KCl, and 5% v/v glycerol at pH 7.5) matched to each sample were recorded and subtracted from the sample data before data averaging. The final experimental scattering curve was calculated using ALMERGE (24) to scale the averaged data sets for each concentration to the highest concentrations (15 mg/mL for M10^short-GCN4 and 12 mg/mL for M10^long-M5CC), merge the data sets, and extrapolate to zero concentration (Fig. S4). The data were analyzed using the ATSAS software suite (25) to determine radii of gyration with AUTORG and distance distributions with DATGNOM (26). The M10^short-GCN4 dummy atom model was generated using DAMMIF (27) and DAMAVER (28) and subsequently processed using the Situs software suite to generate a wireframe envelope representation and align the envelope to the M10^short-GCN4 NMR structure (29). The SBGrid Consortium provided software binaries (23).

Circular dichroism

A Jasco J-1500 spectropolarimeter (Jasco, Oklahoma City, OK) was used for all circular dichroism (CD) spectra and melting curves. CD spectra were collected from 20°C to 98°C with 2°C increments, three spectra per temperature, 1 nm bandwidth. Samples equilibrated for 2 min at each temperature point. Each spectrum was measured from 260 nm to 180 nm, and the melting temperature was analyzed at 222 nm using SigmaPlot 12.0 (SigmaPlot, San Jose, CA) to produce a sigmoidal fit that was replotted using ggplot2 (30).

Results

M10^short-GCN4 is an antiparallel coiled coil

To resolve the myosin-10 controversy we sought structural information on the three coiled-coils (M10^short-GCN4, M10^long-GCN4, and M10^long-M5CC; Figs. 1 A and S2). Using a combination of SAXS and NMR, we first solved the solution structure of M10^short-GCN4. SAXS reveals that M10^short-GCN4 is shaped like a rod (Fig. 2, A and C). The rod is 100 Å long (D_max), which is expected from the length of a 69-residue helical protein. We used DAMMIF (27) to calculate a molecular envelope, shown in Fig. 2 C. Of importance, the envelope lacks the dramatic bends seen in the C-terminal third of the wild type (WT) myosin-10 coiled-coil structure (14) (Fig. 1 C). The DAMMIF envelope encloses a volume with an expected molecular mass of 15 kDa, consistent with a dimer. Moreover, we calculated scattering curves for model monomers, dimers, and trimers using CRYSOL (31), and found that the dimer yields the best fit to the experimental data (Fig. 2 B, reduced χ² of 14, 0.73, and 7, respectively).

Figure 2.

M10^short-GCN4 is a continuous antiparallel coiled coil. (A) SAXS radial distance distribution, P(r), with a specified D_max of 100 Å. The shape of this curve indicates that a high percentage of atoms in M10^short-GCN4 are 15–30 Å apart with the percentage of atoms decreasing as the spacing increases after the maximum. The approximately linear decay after an early peak indicates a rod-like shape, such as a coiled coil (59). (B) M10^short-GCN4 is a dimer. We show the experimental scattering data along with CRYSOL (31) calculated curves for model monomers, dimers, and trimers. The dimer is the final structural model shown in Fig. 2, C and E (the first out of the 10 structures). For the monomer, we deleted one of the chains of the dimer model. For the trimer model, we used the first 69 residues of a long, trimeric coiled coil (PDB: 2WPQ) (60). The dimer yields the best fit (reduced χ² of 14, 0.73, and 7, respectively). (C) NMR structure of M10^short-GCN4, enclosed by the SAXS envelope generated using DAMMIF (gray mesh). Note that the myosin-10 segment (blue) is paired with GCN4 (orange). This structure corresponds to the antiparallel coiled-coil arrangement in Fig. 1D. The lowest energy structural model from a pool of 100 calculated structures is shown. (D) Comparison of the observed NOE restraints grouped by residue (top) with the residue contacts observed in the coordinates (bottom) in (A). Residues in the first chain follow the Bos taurus myosin-10 numbering, whereas residues in the second chain are offset by +1000. The main diagonal shows local NOEs and contacts found in helical structures, whereas the continuity of the minor diagonal shows restraints and contacts caused by the antiparallel coiled coil. The “X” pattern is indicative of an antiparallel coiled coil. (E) Selected residue interactions in M10^short-GCN4. The bottom left panel highlights the “hook” at the C-terminus of the coiled coil where V947 interacts with V888 and I891. Although I891 is in an “a” position of the heptad, V888 is in an “e” position, resulting in overwinding of the coiled-coil superhelix at this site. The bottom right panel illustrates N933, which is a buried “a” position asparagine in GCN4-p1 (N16). In M10^short-GCN4 this asparagine is also buried, but without a buried hydrogen bonding partner. A buried “a”-position lysine residue, K905, is also shown to illustrate that its aliphatic chain supports hydrophobic interactions. To see this figure in color, go online.

We continued with the full NMR solution structure of M10^short-GCN4. The amide region ¹H-¹⁵N HSQC of M10^short-GCN4 contains a single set of well-dispersed peaks, indicating that M10^short-GCN4 forms a symmetric homodimer. We found no sign of additional sets of amide peaks over several months, and concluded that no alternative structures form over this time. We were able to assign backbone and side-chain resonances for all but the first three residues of M10^short-GCN4 (the thrombin cleavage tag), and proceeded with a solution structure derived from ¹³C- and ¹⁵N-separated NOESY distance restraints.

In general, NOESY distance restraints are challenging to assign in homodimeric systems, because any given NOE may report either an intramonomer or an intermonomer distance. Here, the SAXS envelope provides crucial information to resolve this ambiguity. Because the structure has no significant bends, any long-range NOEs (i − j > 5) must come from intermonomer NOEs in an antiparallel geometry. We found a total of 149 of these long-range, intermolecular restraints out of 1561 total NOE distance restraints (Fig. S5; Table 1).

Table 1.

NMR Structure Statistics

	M10short-GCN4a
Completeness of Resonance Assignmentsb
Backbone (%)	96.6c
Side chain (%)	86.4d
Conformationally Restricting Restraintse
Total distance	1561
Intraresidue [i = j]	332
Sequential [|i - j| = 1]	470
Medium range [1 < |i - j| < 5]	610
Long range [|i - j| >= 5]	153
Intrachain restraints	1416
Interchain restraints	149
Dihedral angle restraints	258
Hydrogen bond restraints	228
Disulfide restraints	0
No. of restraints per residue	15.2
(of those, long-range)	1.1
Residual Restraint Violationse
Average no. of distance violations per conformer
0.2–0.5 Å	54.0
> 0.5 Å	0.4 (0.526 max)
Average no. of dihedral angle violations per conformer
1–10°	25.1 (3.9 max)
>10°	0
Model Qualitye
RMSD backbone atoms for ordered residues (Å)	1.3
(all residues) (Å)	1.4
RMSD all heavy atoms for ordered residues (Å)	1.6
(all residues) (Å)	1.7
RMSD bond lengths (Å)	0.011
RMSD bond angles (°)	1.6
MolProbity Ramachandran Statisticse,f
Most favored regions (%)	100.0
Allowed regions (%)	0.0
Disallowed regions (%)	0.0
Global Quality Score (Raw/Z-score)e
Verify3D	−0.01/−7.54
Prosall	0.84/0.79
Procheck (ϕ-ψ)e	1.01/4.29
Procheck (all)e	0.71/4.20
Molprobity clash score	11.87/−0.51
Model Contents
Ordered residue rangesf	883–949
Total no. of residues	138
BMRB accession number	25899
PDB ID	2N9B

RMSD, root mean-square deviation.

^aStructural statistics computed for the ensemble of 10 deposited structures.

^bComputed using CCPN Quality Reports (18) from the expected number of resonances, excluding highly exchangeable protons (N-terminal, Lys, amino and Arg guanido groups, hydroxyls of Ser, Thr, and Tyr), carboxyls of Asp and Glu, and nonprotonated aromatic carbons.

^cUnassigned backbone atoms are from the cleavage tag (881–883).

^dSide-chain assignment statistics for all side-chain atoms (Cα–Cζ), including residues from the cleavage tag^c. Considering only Cα, Cβ, Hα, and Hβ side-chain atoms (including cleavage tag residues), the assignment percentage is 96.4%.

^eCalculated using PSVS version 1.5 (21). Average distance violations were calculated using the sum over r⁻⁶.

^fBased on ordered residue ranges [S(ϕ) + S(ψ) > 1.8]. Chain A and chain B follow same numbering scheme.

We combined these distance restraints with dihedral restraints, backbone hydrogen bonding restraints, noncrystallographic symmetry restraints, and distance symmetry restraints to obtain the solution structure shown in Figs. 2, C and E, and S6. The remarkable feature of this structure is that the two helices are arranged in a homodimeric, antiparallel coiled coil, although M10^short-GCN4 contains the GCN4-p1 leucine zipper sequence (Figs. 3, A and B, and S6). Here, the GCN4-p1 sequence pairs with myosin-10 sequence, but surprisingly not with itself. The crossed and uniform pattern of NOE restraints directly supports the antiparallel coiled coil (Fig. 2 D). Thus, certain sequence contexts flip GCN4-p1, directly challenging the assumption that GCN4-p1 enforces a parallel structure.

Figure 3.

Structural features of M10^short-GCN4 (A) Helical wheel plot of M10^short-GCN4. Note the four favorable and two unfavorable e–g ionic interactions (blue and red dashed lines, respectively). The shifted hydrophobic patches at the J-bends are indicated (red outlines, red arrows). Helical wheels were made using DrawCoil 1.0. (B) Hypothetical helical wheel plot of M10^short-GCN4 in the parallel orientation. The in-register alignment is shown. Note the four favorable and two unfavorable e–g ionic interactions (blue and red dashed lines, respectively), the same as for the antiparallel configuration. Helical wheels were made using DrawCoil 1.0. (C) M10^short-GCN4 bends toward the ends. Helix crossing angles plotted as a function of position along the helix, for all 10 models. Crossing angles are measured for 3–4 residue segments starting at the indicated residue on strand (A), versus a corresponding segment directly across on strand (B). Note the drop in crossing angles toward the end of the structure. The 10 models are the lowest energy structural models from the total pool of 100 calculated structures. (D and E) A coiled-coil distortion on the proximal side of the J-hook. The coiled-coil pitch and radius are shown for M10^short-GCN4 NMR model 1 (D) and model 2 (E), the two lowest energy structural models. Note the peaks indicating increased coiled-coil pitch and radius in the vicinity of residues 896 and 937. Pitch and radius were calculated using the program TWISTER (61). To see this figure in color, go online.

The structure shows the typical coiled-coil interface with a hydrophobic stripe. Residues at the interface are interdigitated, as is common for antiparallel coiled coils (32). The angle between the N-terminal ends of M10^short-GCN4 is 166 ± 8° (SD, over 10 structures), which would tend to splay apart the two myosin-10 motor domains. The C-terminal, CGN4 half of M10^short-GCN4 has a prominent J-shaped bend (Fig. 2 E). This bend allows V947 to sit in a hydrophobic patch between V888 and I891 (Figs. 2 E and 3 C). This hydrophobic patch spans “a” and “e” positions in the heptad repeat, and is rotated away from the normal hydrophobic seam. In the Zhang structure (14), the αB helix overlays the αA helix at this exact location (Fig. 1 C). The J-bend leads to an increase in both the coiled-coil radius and pitch near residues 896 and 937 (Fig. 3, D and E).

In GCN4-p1 structures, residue N16 is a well-known example of a buried polar residue (13, 33). In our structure, this asparagine (N933) also appears in a buried “a” position. However, in our antiparallel structure, N933 sits in a hydrophobic environment near L901 and K905 without a clear hydrogen bonding partner (Fig. 2 E). Consistent with this buried position, the side-chain amide nitrogen, Nδ2, is an upfield chemical shift outlier (105.004 ppm).

Several factors contribute to coiled-coil orientation (32, 34, 35, 36). Charged side chains at “e” and “g” positions can form interhelical salt bridges, but M10^short-GCN4 would have similar salt bridge patterns in both parallel and antiparallel forms (Fig. 3, A and B). The presence of β-branched residues within the hydrophobic seam also favors antiparallel orientations, especially when paired with β-unbranched residues from the partner strand (37). There are 10 of these β-branched residues in the “a” positions of M10^short-GCN4 (Fig. 3 A). Six out of 10 of these interactions are among the most favorable for antiparallel orientations (four IL pairs, and two VL pairs) (35). Although GCN4-p1 can be mutated to form antiparallel tetramers (38, 39), M10^short-GCN4 is the only known antiparallel and dimeric structure with WT GCN4-p1 sequence.

The longer coiled-coil designs are partially folded

Even though M10^short-GCN4 forms antiparallel dimers, the longer M10^long-M5CC and M10^long-GCN4 have different sequence contexts and might form entirely different structures. We find that both M10^long-GCN4 and M10^long-M5CC have unusually large hydrodynamic radii, suggesting that they may form larger complexes at high concentration (Fig. 4 A). We compared amide-region ¹⁵N-HSQC spectra of M10^short-GCN4 (Fig. 4 B) and M10^long-GCN4 (Fig. 4 C), and observed fewer amide peaks than expected. The poor spectral dispersion and considerable peak overlap for M10^long-GCN4 prevents the assignment of backbone amide peaks and is a direct indicator of conformational dynamics on the NMR timescale. Moreover, M10^long-GCN4 HSQC spectra are strongly temperature dependent from 25–45°C and show signs of additional flexibility at the higher temperatures (Fig. S7, A–C). A comparison of M10^long-GCN4 peak locations with those of the WT myosin-10 coiled coil, GCN4-p1 (40), and M10^short-GCN4 shows no overlapping peaks, suggesting that M10^long-GCN4 is structurally dissimilar to all three.

Figure 4.

M10^long-GCN4 is dynamic and lacks an all-or-none folding transition. (A) Preparative gel filtration chromatograms from the final stage of purification of M10^short-GCN4, M10^long-GCN4, and M10^long-M5CC. These three proteins elute at 23, 71, and 130 kD, respectively. All three elute at larger masses than predicted from globular standards, as expected for rod-like molecules. However, M10^long-GCN4 elutes at a much greater MW than M10^short-GCN4, even though these two constructs are of similar length. (B) The amide-region ¹H-¹⁵N HSQC spectrum of M10^short-GCN4 illustrates good peak dispersion. (C) The corresponding M10^long-GCN4 HSQC has poor dispersion and has fewer peaks than backbone amides, indicative of conformational dynamics. (D) The M10^short-GCN4 CD melting curve at 222 nm shows a cooperative unfolding transition. (E) The corresponding M10^long-GCN4 CD shows gradual helical fraying without evidence of a cooperative transition. To see this figure in color, go online.

To compare the secondary structure stability of M10^short-GCN4 and M10^long-GCN4, we used CD thermal denaturation. M10^short-GCN4 unfolds cooperatively at increasing temperature, with a sigmoidal response in mean residue ellipticity at 222 nm and a T_m of 48.6°C (Figs. 4 D and S7 D). However, M10^long-GCN4 gradually melts without cooperativity, an additional indicator of a frustrated, flexible structure (Figs. 4 E and S7 E).

Interestingly, the M10^long-GCN4 must still self-assemble, as assembly is a prerequisite for processive myosin motility. We note that M10^long-GCN4 is 61% helical (Fig. S7 E) and contains a heptad repeat that would tend to form a continuous hydrophobic seam, even though it lacks folding cooperativity. Note that certain molten globule protein states can form specific intra- or intermolecular interactions, despite their considerable flexibility (41). Takagi et al. observed HMM dimers containing M10^long-GCN4 in their electron microscopy (EM) images (11). However, these Takagi EM images also suggest flexibility in their construct. The two motor domains sample a wide range of angles, similar to myosin-5, and are not splayed apart at 180° (11). Note that interactions with the carbon grid might affect the orientation distribution; therefore, some caution is needed when inferring flexibility from EM images.

At 243 residues per monomer, M10^long-M5CC is too large for NMR, so we instead used SAXS to characterize its structure. Comparing the M10^short-GCN4 and M10^long-M5CC Guinier plots (Fig. 5, A and B), M10^long-M5CC is either aggregated or quite large with an R_g of 126 Å. The radial distribution function shows that the maximum dimension of M10^long-M5CC is at least 500 Å. This distance is much longer than expected from a single, parallel coiled coil with the length of the M10^long-M5CC sequence (Fig. 5, C and D). To explain this observation, we propose that the long, stable myosin-5 sequence forms a parallel coiled coil, which prevents the N-terminal myosin-10 segment from folding. Thus, M10^long-M5CC is in a parallel, frustrated state (Fig. 1 D). At the high concentrations used in SAXS, we suspect that antiparallel segments from different dimers can pair to generate oligomers with a longer end-to-end distance. At the low concentrations used in single-molecule experiments (9), the myosin-10 segments are likely frustrated and have conformational dynamics much like M10^long-GCN4.

Figure 5.

M10^long-M5CC aggregates or assembles. (A) The SAXS Guinier region of M10^short-GCN4 is linear, and yields a R_g of 27 Å, consistent with the 30 Å R_g expected for a rod with a length of 100 Å. (B) The Guinier region of M10^long-M5CC is unusually narrow, yielding an R_g of 126 Å. (C) The M10^long-M5CC radial distribution function, P(r), calculated using a maximum distance, D_max, of 360 Å. This length corresponds to a rod model where the entire M10^long-M5CC sequence is a parallel coiled coil. Note the abrupt decay at high r, indicative of an underestimated D_max. (D) The best D_max of M10^long-M5CC is 500 Å, with a real-space R_g of 138 Å. Thus, M10^long-M5CC likely forms an oligomeric structure that is longer than a single dimer under these conditions.

Discussion

We have found that fused parallel and antiparallel coiled coils can adopt surprising and unpredictable structures, depending on their exact sequences. To determine the prevalence of orientation switching in sequences that contain GCN4-p1, we performed a BLAST search against the Protein Data Bank (PDB) using GCN4-p1 as the query sequence. We found three reports of antiparallel forms of GCN4 (38, 42, 43). However, these examples differ because all are mutants of GCN4-p1 and all are trimers or tetramers. The M10^short-GCN4 is the first reported, to our knowledge, structure of orientation switching in a dimer with a WT GCN4-p1 sequence. For this fused parallel and antiparallel coiled coil, it is energetically favorable to force the GCN4 sequence to be antiparallel, rather than to force the myosin-10 sequence to be parallel.

Why does M10^short-GCN4 fold, whereas M10^long-GCN4 and M10^long-M5CC each have structural issues? The problem may be that M10^long-GCN4 and M10^long-M5CC designs include the full myosin-10 bend (Fig. 1, A and C), with the skips in the heptad repeat. Although Zhang proposed that the longer M10^long-M5CC could fold in the genie-bottle form with an intact, antiparallel myosin-10 coiled coil (14), the linkers between the parallel and antiparallel segments are likely too short. Such extended linkers can be challenging to design (44). In contrast, the M10^short-GCN4 design has a continuous heptad repeat that accommodates a straight coiled coil, albeit an antiparallel one. Although the dimerization affinity of GCN4-p1 is much higher than the myosin-10 coiled-coil (K_D of 26 nM vs. 590 nM) (14, 45), affinity alone does not determine the orientation. Indeed, coiled-coil orientation and oligomerization state are difficult to predict due to a combinatorial explosion of possible interactions (46).

The M10CC^short-GCN4 forms an antiparallel coiled coil through a structural competition where the myosin-10 segment overrules the GCN4-p1 segment. Certain natural and engineered protein functions exploit such structural competition. For example, the dramatic rearrangements in the core HA₂ domain of hemagglutinin harness structural competition to drive membrane fusion (47, 48). Moreover, a Zn²⁺ transporter called the “Rocker” uses structural competition to allow only one of two metal-binding sites to be occupied simultaneously as part of its design (49).

How might an antiparallel coiled coil affect the stepping of myosin-10? Zhang proposed that the antiparallel coiled coil extends the reach of myosin-10, allowing both motor domains to contact a single actin filament. In this model, the coiled coil acts as a shoulder that spaces two arms (14). However, our earlier work refutes this proposal. We can disrupt selectivity of myosin-10 by inserting a swivel (a flexible glycine-serine linker) at the N-terminus of the coiled coil (7). In the Zhang model, the ends of the coiled coil are intrinsically flexible, analogous to the rotator cuff of the shoulder. If the Zhang model were correct, additional flexibility from our linker insertion should have had no effect on myosin-10 stepping.

We propose an alternative model, extending our original proposal that myosin-10 walks by straddling two actin filaments in a bundle. The key feature is that the two single α-helix (SAH) domain helices project from the coiled coil without breaks in the helical structure, on average. In an actin filament bundle, myosin-10 can reach adjacent actin filaments with a gradual SAH domain bend (Fig. 6 A) (50, 51). However, myosin-10 cannot place both motor domains on a single filament without considerable strain (Fig. 6 B). Such strain would increase the likelihood of detaching before successfully completing a step on a single filament. Antiparallel coiled coils and oriented SAH domains would therefore favor walking on bundles. Likewise, a dynamic dimerization region breaks this bundle selection mechanism (Fig. 6 C).

Figure 6.

A model for bundle selection using antiparallel coiled coils. (A) Myosin-10 straddles two different actin filaments in a fascin-actin bundle. Starting and ending myosin-10 configurations for one step are shown, with the starting state in transparent red and the ending state in solid red. The most common target actin-binding sites are highlighted in blue. The moderate bend of the SAH:coiled-coil:SAH elements (thin curved line) allow both motor domains to engage separate actin filaments. (B) Myosin 10 on a single actin filament, with one bound motor domain. The antiparallel coiled coil forces the free motor domain to project away from the only available actin filament. Moderate flexibility, illustrated using transparent motors, inhibits rebinding to actin. Processivity is greatly reduced with significantly shorter runs. (C) Additional flexibility at the coiled coil allows the free motor domain to reach the actin, and enables processive stepping. Flexible HMM constructs include those containing M10^long-GCN4, M10^long-M5CC, or the swivel (7). Although the dimerization domain is likely dynamic, for simplicity it is illustrated here with a short parallel segment. Note that all myosin motor domains are shown with arbitrary lever arm orientations as these are currently unknown. To see this figure in color, go online.

Single-molecule total internal reflection fluorescence motility assays can select for populations of walking myosins. Therefore, the prior studies on myosin-10s with M10^long-GCN4 or M10^long-M5CC may have detected populations that were structured and oriented to allow motility on single filaments. However, for these populations to remain processive for 40–60 steps, the structural states would have to exchange slowly, on timescales longer than several seconds. If this slow exchange were happening, we would expect to see multiple full sets of amide peaks in the NMR of M10^long-GCN4, one set for each structural state. Instead, we see a spectrum that is typical for a dynamic protein with little tertiary structure.

A comparison of the WT myosin-10 coiled-coil structure with our M10^short-GCN4 structure finds that both orientations are similar. The N-terminal ends of both structures project at nearly the same large angle (Zhang: 161 ± 7°, M10^short-GCN4: 166 ± 8°, ±SD over top 10 structures, Fig. 7). Thus, we expect that this bundle selection mechanism would operate in WT myosin-10 as well. The main structural difference is that the M10^short-GCN4 is about twice as long as the WT myosin-10 coiled coil (100 Å vs. 45 Å, respectively, Fig. 7). Although we cannot rule out that a particular coiled-coil length is critical for actin bundle selection, we suspect that orientation is much more important. For example, when we replace the myosin-5 coiled coil with the myosin-10 SAH + M10^short-GCN4 sequences, we make a bundle-selective myosin-5 (7). This selective myosin-5 has six additional IQ domains compared to myosin-10, and is over 200 Å longer.

Figure 7.

Structural comparison of M10^short-GCN4 (top) and the WT myosin-10 coiled coil (bottom) (14). The N-terminal ends (back helices) of the M10^short-GCN4 span 166 ± 8°, whereas the corresponding N-terminal ends (back helices) of the Zhang structure span 161 ± 7° (±SD over top 10 structures). Thus, the orientation of the important upstream SAH, IQ, and motor domains should be nearly identical. However, M10^short-GCN4 is about twice as long as the Zhang coiled coil. Boxes indicate the hydrophobic patches on the back helices that are illustrated on the helical wheel plot in Fig. 3A and in Fig. 2E. These patches are covered by the J-hooks in M10^short-GCN4, and the αB helices in the Zhang structure. To see this figure in color, go online.

The approach of fusing GCN4-p1 and its mutants to assemble proteins is common (52, 53, 54, 55, 56). In many of these applications, dimerization is the primary consideration and the final orientation of the dimer is only of secondary importance. However, in certain proteins the orientation will strongly affect function. With motor proteins in particular, dimerization domains can profoundly affect activity. Our results provide cautionary evidence that coiled-coil sequences containing GCN4-p1 may be either parallel or antiparallel, and that orientation must be established through independent structural work.

Author Contributions

K.C.V., Y.X., and R.S.R. designed research; K.C.V. and Y.X. performed research; K.C.V., Y.X., and R.S.R. contributed analytic tools; K.C.V. and R.S.R. analyzed data and wrote the article.

Editor: E. Ostap.

Accession Numbers

The accession number for the M10^short-GCN4 data reported in this paper is PDB: 2N9B. The accession number for the NMR data reported in this paper is BMRB: 25899.