Structural Analysis of Human Cofilin 2/Filamentous Actin Assemblies: Atomic-Resolution Insights from Magic Angle Spinning NMR Spectroscopy

Cellular actin dynamics is an essential element of numerous cellular processes, such as cell motility, cell division and endocytosis. Actin’s involvement in these processes is mediated by many actin-binding proteins, among which the cofilin family plays unique and essential role in accelerating actin treadmilling in filamentous actin (F-actin) in a nucleotide-state dependent manner. Cofilin preferentially interacts with older filaments by recognizing time-dependent changes in F-actin structure associated with the hydrolysis of ATP and release of inorganic phosphate (Pi) from the nucleotide cleft of actin. The structure of cofilin on F-actin and the details of the intermolecular interface remain poorly understood at atomic resolution. Here we report atomic-level characterization by magic angle spinning (MAS) NMR of the muscle isoform of human cofilin 2 (CFL2) bound to F-actin. We demonstrate that resonance assignments for the majority of atoms are readily accomplished and we derive the intermolecular interface between CFL2 and F-actin. The MAS NMR approach reported here establishes the foundation for atomic-resolution characterization of a broad range of actin-associated proteins bound to F-actin.

Budding yeast and human cofilins share ~38% identity and 59% homology, while the three human isoforms of ADF/cofilins share 70-81% identity and 83-89% homology. The structure of ADF/cofilins is highly conserved across all eukaryotic species. The core of ADF/cofilins is comprised of a five-stranded mixed β -sheet and is surrounded by five peripheral α -helices. The secondary structure elements are positioned in the following order: α 1-α 2-β 1-β 2-α 3-β 3-β 4-α 4-β 5-α 5-β 6. The C-terminus of cofilins is folded in a compact β 6-strand, which distinguishes it from a less tightly packed C-terminus of ADF 12 ; this difference is translated to a lower affinity of the latter to F-actin. Structural and mutagenesis studies as well as mapping with synchrotron irradiation and chemical cross-linking [12][13][14][15][16] converged to realization that ADF/cofilins interact with actin via two major areas called G/F-and F-binding sites. The former is involved in binding to both monomeric and filamentous actin and is mainly composed of the N-terminal residues, and the kinked α 4-helix. The N-terminus contains a major regulatory residue Ser3, whose phosphorylation by LIM kinase 17 strongly inhibits binding of cofilins to G-and F-actin, as do point mutations at the N-terminus and in the α 4-helix 12,18,19 . The F-actin binding site does not interact with G-actin, but is essential for binding to actin filaments; respectively, mutations in this region (e.g. K96Q 12 ) do not affect binding to monomeric actin, but block the ability of cofilin to affect F-actin dynamics. The unique role of ADF/coflins in actin dynamics is directly linked to its ability to induce dramatic structural perturbations in F-actin. Cofilin binding to F-actin promotes relative reorientation of the actin subunits leading to a change in a helical twist of the filament 20 , and an increase in its torsional and bending flexibility 21,22 . Cofilin modulates filament interaction with actin binding proteins via direct competition/cooperation, via allosteric influences, or both [23][24][25][26] . Recent cryo-EM reconstruction studies 13 revealed that reorientation of actin subunits is required to avoid steric clashes between the α 1-and α 4-helices of cofilin with actin subdomains 1 and 2 (SD1 and SD2), respectively. Thus, binding of cofilin causes a rotation of the outer domain of actin (comprised of SD1, 2) and weakens the interface between SD1 and SD2 of the longitudinally adjacent actin subunits. While these studies yielded key insights, advanced understanding of cofilin-actin interaction and the molecular mechanism of cofilin-based actin remodeling requires atomic level structural information on these complexes, which is currently lacking.
In this report, we present atomic-resolution structural analysis of the muscle isoform of human cofilin (CFL2), bound to the α -skeletal isoform of mammalian F-actin, by magic angle spinning (MAS) NMR spectroscopy. The three isoforms of mammalian cofilins are indispensable and vary in distribution in normal and diseased tissues 3,27,28 , abilities to support cellular activities 29,30 , and in their affinities for G-and F-actin 3,9,31,32 . Among these, only CFL2, the muscle-specific isoform of ADF/cofilins, was characterized in complex with F-actin by cryo-EM reconstruction at 9 Å resolution 13 , and therefore, was selected for this study.
MAS NMR spectroscopy, which is emerging as a mainstream structural biology technique, has been applied to studies into cytoskeleton protein assemblies, such as determination of atomic-resolution structure and dynamics of CAP-Gly domain assembled with polymerized microtubules 33,34 , and an investigation of the polymorphism of myelin basic protein associated with actin microfilaments 35 . We have determined the chemical shifts and the secondary structure for the majority of cofilin residues. Using chemical shift perturbation analysis in conjunction with double-REDOR (dREDOR) filtered methods 36 , we have delineated the cofilin residues forming the interfaces with F-actin as well as residues that are affected allosterically. The outstanding spectral resolution enabled characterization of the cofilin/actin interactions with unprecedented level of detail. The study presented here opens doors for atomic-resolution structural characterization of assemblies formed by F-actin with cofilin/ADF family and other ABPs.

Results And Discussion
Resonance assignments of cofilin in complex with actin filaments. The negatively stained TEM image of the NMR sample containing U-13 C, 15 N human CFL2 in complex with F-actin is shown in Fig. 1c. Formation of CFL2/F-actin complex was confirmed by cosedimentation and SDS-PAGE (Fig. 1d). To corroborate that MAS NMR conditions do not interfere with the sample morphology, TEM images were collected prior to and after the NMR experiments. Spinning of the samples for extended periods of time does not appear to have any effect on the sample morphology. The samples of CFL2/F-actin complexes yield outstanding-resolution MAS NMR spectra, as shown in Fig. 2. For resonance assignments of cofilin in complex with actin filaments, a combination of 2D and 3D homo-and heteronuclear correlation MAS NMR spectra were acquired at 19.96 T Backbone walk for the stretch of residues D9-N16 obtained from 3D NCACX spectra (green contours) and 3D NCOCX spectra (black contours). (c) Human CFL2 sequence with secondary structure of CFL2 determined by TALOS+ , where blue arrows represent β -strands, green boxes represent α -helices and yellow box represents 3 10 helix Colored in yellow and black are residues that are assigned and unassigned (not assigned a single atom), respectively. Spectra were processed with 30 degree sinebell and Lorentizan-to-Gaussian apodization. The first contour is set at 5x the noise level.
SCientifiC REPORTS | 7:44506 | DOI: 10.1038/srep44506 using non-uniform sampling (NUS). As shown in Fig. 2a,b and Table S1 (Supplementary Information), remarkably high spectral resolution of 2D and 3D spectra permitted de novo resonance assignments of 111 out of 166 residues. Figure 2b displays a backbone walk from D9-N16 using 3D NUS NCOCX and NCACX datasets. The majority of backbone resonances are present in the spectra with the exception of termini residues (M1, P165, L166), several sequence stretches in loop regions spanning residues K30-K34, V72-D79, G130-Q136, and random individual residues throughout the sequence ( Fig. 2c and Table S1). These assignments were also corroborated by comparison with solution shifts of free cofilin.

Intermolecular interface and allosteric changes in cofilin upon binding to F-actin.
To determine the intermolecular interface of CFL2 in complex with F-actin, we have pursued two approaches: (i) chemical shift perturbation analysis, and (ii) dREDOR-filter-based experiments allowing for a direct determination of intermolecular dipolar contacts between cofilin and actin filament. This combined analysis yields information on the intermolecular interface and on residues experiencing allosteric changes as the result of the complex formation.
The above strong chemical shift perturbations indicate that the corresponding residues either comprise the intermolecular interface with F-actin or undergo allosteric conformational changes upon formation of the complex. To discriminate between those possibilities and identify cofilin residues located at the interface with F-actin, we employed a dREDOR filtered approach 33 . In this method, simultaneous 1 H-13 C/ 1 H-15 N REDOR filter dephases all cofilin protons that are directly bonded to either 13 C or 15 N atoms (i.e., 1 H belonging to the U-13 C, 15 N-cofilin). The remaining protons, belonging to the interface regions of unlabeled actin, are then used to transfer magnetization to cofilin through 1 H-15 N or 1 H-13 C cross polarization across the intermolecular interface followed by HETCOR or CORD mixing. The resulting spectrum contains specific information about the residues forming intermolecular interface. We note that dREDOR based experiment is the only approach to identify intermolecular interfaces when chemical shift perturbations are small, such as in cofilin/actin complex here or CAP-Gly/microtubule complex studied by us previously 33 . To perform residue assignments of cofilin residues at the interface, this experiment is implemented as a 2D dREDOR-CORD sequence (Fig. 4a).

G/F-actin binding sites on cofilin.
The dREDOR-based measurements reveal both the G/F-and F-binding sites on the cofilin surface (representing binding to subunits "n" and "n + 2" in the filament, respectively) to previously unprecedented level of detail. On the basis of the dREDOR-CORD experiment, we confirmed that the G/F-binding site includes the N-terminus (residues S3, G4 and V6) and the N-terminal half of the bent α 4-helix (residues M115 and I116) preceding the kink. Interestingly, the entire actin-binding surface of the α 4-helix appears to be shifted towards its C-terminus as compared to that observed in the crystal structure of  14 , thus far the only atomic resolution structure of actin-cofilin complex available. Indeed, the N-terminal residues of the α 4-helix that constitute the essential part of the G/F-binding site in the C-twinfilin structure (266-IRER-269 corresponding to 111-LKSK-114 in human CFL2) are not present in the dREDOR-CORD spectra. On the other hand, the data indicate that residues A123, I124, K127, T129 at the C-terminal part of the kinked α 4-helix, which are not at the G-actin/C-twinfilin interface, are part of the interface in the CFL2/F-actin complex. Residues located in the corresponding part of the α 4-helix of budding yeast cofilin were found by mutagenesis to be essential for cofilin function in vivo 18 and were recognized to be affected upon binding to G-actin in synchrotron oxidation experiments 15 and in solution NMR studies 12 , see Fig. 5. The fact that strong signals corresponding to these residues are found in dREDOR spectra suggests that actin is directly involved in interaction with the terminal part of the α 4-helix. It remains to be determined whether the differences observed in the various studies reflect variations in actin-binding modes between the ADF-homology domains of C-twinfilin and cofilin, or between interaction of cofilins with G-vs F-actin, or even between binding of cofilin to ATP-(the X-ray structure) vs ADP-actin (present work). It is also possible that these additional residues contribute to a new F-actin binding site, by being possibly involved in binding to the D-loop of a longitudinally adjacent actin subunit ("n + 2"), as it can be speculated based on the proximity of these elements in the cryo-EM reconstruction of F-actin with (a) dREDOR-CORD spectra for human CFL2 in complex with F-actin. The residues constituting the corresponding intermolecular CFL2/F-actin interfaces were mapped onto the CFL1 structure (PDB 1Q8G) and are shown in purple in (b). Spectra were processed with 30 degree sinebell and Lorentizan-to-Gaussian apodization. The first contour is set at 5x the noise level.

F-actin binding sites on cofilin.
To date, the most comprehensive data on the F-site composition originated from a restrained refinement model obtained by fitting atomic-resolution structures of CFL1 (PDB 1Q8G) and actin (PDB 2BTF) to a 9 Å resolution cryo-EM map density of human CFL2 with rabbit skeletal actin 13 . The dREDOR-CORD spectra (Fig. 6) are in excellent agreement with the cryo-EM reconstruction data, and reveal the three major patches of the F-site (to subunit "n + 2" in F-actin) (Fig. 6). The first patch is defined by residues 19-21 (also recognized in the cryo-EM study) and 24-26. It is reasonable to speculate that the latter patch can be involved in binding to the N-terminus of actin as these elements are proximal to each other in the cryo-EM reconstruction 13 . The loop comprised by residues 24-32 is present in vertebrates, from fish to human, but not in drosophila, yeast, or plant cofilins. This loop contains a nuclear localization signal (NLS) essential for active translocation of G-actin-cofilin complexes to the nucleus 37 . Involvement of the 24-32 loop in interaction with F-, but not G-actin, suggests a possible mechanism of discriminative recognition of free cofilin and actin-cofilin complexes by nuclear importins. Furthermore, phosphorylation of S23 and/or S24 by PKCα reduces ability of cofilin to bind F-actin and to modify actin dynamics both in vitro and in vivo 38 . Given a highly acidic nature of actin's N-terminus, it is reasonable to suggest an electrostatic repulsive mechanism for such inhibition. Therefore, this site represents regulation at the F-actin level, as opposed to the G/F-level regulation via phosphorylation at S3.
The second patch of the F-site is composed of residues 91, 93-96, and 99, 100 (residues 94-98 in the cryo-EM study) within the loop connecting β 4-β 5 strands and the N-terminal half of β 5-strand. The third patch encompasses residues 153, 157-160, and 163 at the C-terminus of cofilin (defined as residues 154-158 by Galkin et al. 13 .). It was speculated that overall tighter folding and stronger binding of this regions to F-actin defines the major structural and functional differences between cofilin and ADF 12 . In addition to these well-defined patches, dREDOR spectra revealed residues T63, T69, S70, which are located at the surface opposite to actin binding sites. As illustrated in Fig. 3, the majority of the residues experiencing large CSPs are either present in the dREDOR spectra or are directly neighboring them corroborating the finding that these residues are located at the interface with F-actin. Among few cofilin residues with large chemical shift perturbations that are not in proximity of residues in dREDOR spectra, V57 (~8 ppm), D141, V143 (4-6 ppm), C80 (~2 ppm), and T88 (> 2 ppm), are particularly notable (see Fig. 3). Their unusually large CSPs correlate with localization in loops (V57, C80 and T88) and, therefore, likely reflect flexibility of the corresponding loop regions, or with localization at the actin-cofilin interface (D141, I143) not identified by the dREDOR spectra. Indeed, D141 and D142 of CFL2 correspond to E141 and E142 in CFL1 sequence. This difference was recently implicated as a likely source of different affinities of the two cofilins to ATP-actin 9 . In addition, high CSP of C80 likely reflects changes in the oxidation state of this residue (reduced in free cofilin and oxidized in complex with actin, according to the 13 C α /C β chemical shifts 39 ). Additional insights into the nature of the changes described by the large chemical shift perturbations will be gained from the atomic-resolution MAS NMR structure of CFL2 bound to actin, which is forthcoming.

Conclusions
We reported atomic-resolution structural investigation of the human CFL2 complex with F-actin, by MAS NMR spectroscopy. Using a combination of chemical shift perturbation analysis and dREDOR-based methods, we have determined the cofilin residues comprising the interfaces with F-actin as well as residues affected allosterically. The outstanding spectral resolution enabled characterization of the cofilin/actin interactions with unprecedented level of detail. Broadly, MAS NMR approach reported here is applicable to the analysis of actin-binding proteins bound to filamentous actin that otherwise are not amenable to atomic-resolution studies through other current techniques.

Methods
Materials. 15 NH 4 Cl and U-13 C 6 glucose were purchased from Cambridge Laboratories, Inc. Common chemicals were purchased from Fisher Scientific or Sigma-Aldrich.
Preparation of F-actin. Skeletal muscle G-actin was prepared from acetone powder of rabbit skeletal muscle (Pel-Freeze Biologicals) as previously described 41 15 N-CFL2 and F-actin were mixed at a 1:1.2 molar ratio and centrifuged at 90,000 rpm (RCF 435,400 g) at 4 °C for 1 hour in a TLA 120.2 rotor using a Beckman Coulter Optima MAX-XP ultracentrifuge following overnight incubation on ice. The gel-like pellet was transferred into a 1.9 mm Bruker rotor. 14.3 mg of hydrated U-13 C, 15 N-cofilin/actin complexes containing an estimated 3 mg of isotopically labeled cofilin were packed into 1.9 mm Bruker rotors.
Transmission Electron Microscopy. The U-13 C, 15 N-cofilin/actin sample morphology was verified by transmission electron microscopy (TEM). Actin samples were stained with uranyl acetate (5% w/v), deposited on 400 mesh, formvar/carbon-coated copper grids, and dried. The TEM images were acquired by a Zeiss Libra 120 transmission electron microscope operating at 120 kV. Solution NMR Spectroscopy. Solution NMR spectra were acquired on a 14.1 T ( 1 H Larmor frequency of 600.1 MHz) Bruker AV spectrometer using a triple-resonance inverse detection (TXI) probe. All spectra of U-13 C, 15 N-cofilin were recorded at 298 K. Backbone and C β resonance assignments of U-13 C, 15 N cofilin were carried out using heteronuclear 2D 1 H-15 N HSQC and 3D HNCACB, HNCA, HNCO, HNCACO at 298 K.

MAS NMR Spectroscopy.
MAS NMR spectra were acquired on a 19.96 T Bruker AVIII instrument using a 1.9 mm HCN probe. The Larmor frequencies were 850.4 MHz ( 1 H), 213.8 MHz ( 13 C) and 86.2 MHz ( 15 N). The MAS frequency was 14 kHz, controlled to within ± 10 Hz by a Bruker MAS controller. The temperature was calibrated using KBr as the temperature sensor. The actual temperature at the sample was 273 K and maintained to within ± 0.1 °C using a Bruker temperature controller. 13 C and 15 N chemical shifts were referenced with respect to the external standards adamantane and NH 4 Cl, respectively. Dipolar-based 2D and 3D NCACX and NCOCX experiments were acquired using non-uniform sampling (NUS). The random exponentially weighted NUS schedules were employed, see Supplementary Information 42,43 . The 3D spectra were acquired with 25% NUS using 48 complex points in the t 1 and t 2 indirect dimensions, with maximum evolution times of 3.4 ms and 6.9 ms for 13 C and 15 N, respectively. The spectra were processed using the MINT reconstruction protocol 42 . The signal-to-noise ratio (SNR) of the 2D NCACX and NCOCX were 14 and 13, respectively. The SNR for the 2D 13 C-13 C CORD is 15. The first contour level was set to 5X the noise level for all spectra. The uncertainties in the solid-state chemical shifts are ± 0.3 ppm. The typical 90° pulse lengths were 2.75 μ s, 2.95 μ s for 13 C, and 3.3 μ s for 15 N. The 1 H-13 C and 1 H-15 N CP employed a linear amplitude ramp for 80-100%: the 1 H RF field was 91 kHz; and the center of the ramp on the 13 C or 15 N was Hartmann-Hahn matched to the first spinning sideband. In 2D and 3D NCACX experiments, the RF field strengths were 64.9 kHz, 84.7 kHz and 91 kHz for 15 N, 13 C and 1 H channels, respectively. The DARR mixing sequence was applied to the 1 H channel and the DARR mixing time was 50 ms. The 1 H decoupling powers were 90-100 kHz during acquisition and evolution periods in all experiments.
For 2D 13 C-13 C CORD correlation experiments 44 , the typical 90° pulse lengths were 2.55 μ s for 1 H, and 2.3 μ s for 13 C. The 1 H-13 C CP employed a tangent amplitude ramp of 80-100%, the 1 H RF field was 75 kHz, and the center of the ramp of the 13 C Hartmann-Han matched the first spinning sideband. The RF field on the 1 H channel was matched to the MAS frequency (14 kHz) and one half of it (7 kHz) during the 50 ms mixing time. The typical decoupling power was 90-100 kHz during the acquisition and evolution. The spectral width was 299.75 ω 2 and 213.83 in ω 1 with the carrier frequency set to 96.2 ppm.
The double-REDOR (dREDOR) filtered experiments on U-13 C, 15 N-CFL2 bound to F-actin employed simultaneous 1 H-13 C/ 1 H-15 N REDOR dephasing periods of 714 μ s, to eliminate signals from 1 H (all protons in CFL2) directly bound to 13 C and 15 N 33 . The CP contact time was 5 ms and the CORD mixing time was 50 ms. To rule out the possibility that certain signals in the dREDOR based spectra may be artifacts arising from dynamic residues due to insufficient suppression of 1 H spin polarization, control experiments were conducted as described in our report 45 .