In silico conformational dynamics of the α-actinin-2 actin-binding domain upon phosphorylation

Abstract

In primary cardiomyocyte cell cultures, α-actinin-2 phosphorylation at the actin-binding domain (ABD) increases with mechanical stress, facilitating adaptation to varying forces. Nevertheless, it is unknown whether these phosphorylation sites in α-actinin-2 have structural consequences that could explain differential binding to F-actin and influence sarcomere stabilization and assembly. This study aims to understand the mechanisms by which α-actinin-2 regulates the assembly dynamics of sarcomeres in the heart. To investigate phosphorylation-specific modifications at the α-actinin-2 ABD, structural modeling was conducted on phosphorylation sites T43, S50, S147, and T237 alongside their phosphomimetic counterparts T43D, S50D, S147D, and T237D. We quantified conformational changes at the ABD using complementary AlphaFold 3-generated models and molecular dynamics (MD) simulations of the phosphomimetic variants. AlphaFold 3 modeling of phosphorylation and pseudo-phosphorylation sites showed an increase in the distance between the centers of mass of the two calponin homology domains (CH1 and CH2), and a decrease in torsion angles, opening the α-actinin-2 ABD. These structural changes correlated with more favorable electrostatic interaction energies (ΔH_elec) toward actin. C_α RMSD values for the superimposed α-actinin-2 WT and phosphorylated, as well as pseudo-phosphorylated variants, demonstrate the alignment of the CH1 domains, but straightening of the loop region, separating the CH1 and CH2 domains. Phosphorylation and pseudo-phosphorylation of all the studied residues increase the net negative electrostatic potential at the CH2 domain while causing a net positive to net negative transition at the CH1 domain. MD simulations show that all the phosphomimetic mutations increased the spread in CH1-CH2 torsional and distance-dependent conformations with S147D exhibiting the largest spread followed by T237D and T43D. These results imply that α-actinin-2 phosphorylation affects the structural stability of the closed conformation of the ABD, suggesting that phosphorylation could promote the open state of α-actinin-2 ABD and facilitate its engagement with actin.

Graphical Abstract:

INTRODUCTION

Cells respond to mechanical stimuli through mechanotransduction, a process in which mechanosensitive sarcomeres convert mechanical forces into biochemical signals, resulting in cellular responses such as cytoskeletal remodeling. This includes both biochemical signaling pathways and neurohumoral signaling, upon which cells sense and respond to their physical surroundings by modifying their structure and function (1). The sarcomeres, composed of repeating units of thin filaments, which comprise primarily of actin, and myosin-containing thick filaments, undergo changes in response to mechanical forces, transmitted through the α-actinin protein (2–4). α-actinin is an actin crosslinking protein crucial for maintaining cytoskeletal integrity and cell movement. It is found throughout the cell, including both the cellular cytoskeleton and cortical cytoskeleton, and is especially important at cell-cell and cell-matrix contact sites, as well as in stress fiber-dense regions (5, 6).

Out of the four α-actinin isoforms (ACTN1, ACTN2, ACTN3, ACTN4 genes), α-actinin-2 and α-actinin-3 are muscle-specific Ca²⁺-insensitive crosslinkers, meaning that Ca²⁺ is not required to trigger the assembly of α-actinin-2 and α-actinin-3 with other cytoskeletal proteins (7,8). Both, α-actinin-2 and α-actinin-3, are localized to the Z-disk of sarcomeres in skeletal and smooth muscles, where they anchor actin filaments and play a role in thin filament organization and interaction with other Z-line proteins, but α-actinin-2 is the only isoform that is expressed in the striated cardiac muscles (9,10). Interestingly, α-actinin-2 and α-actinin-3 can be found in the Z-discs, as well as in the Z-bodies of striated and non-striated muscle cells, suggesting their role in sarcomere assembly and striated/non-striated muscle development (11). In cardiac muscle cells, α-actinin-2 functions as an antiparallel homodimer to crosslink actin filaments at the Z-discs, anchoring and organizing actin thin filaments from neighboring sarcomeres (4).This crosslinking is achieved through its actin-binding domain (ABD), which is composed of two calponin homology domains (CH) in tandem (9, 12).

All human α-actinin isoforms share a common, highly conserved primary and tertiary structure with 84% sequence identify for the non-muscle and 80% for the muscle isoforms (13).The N-terminal ABDs, that contain two consecutive CH domains, are positioned at either end of the α-actinin homodimer stretch, allowing α-actinin-2 to cluster and cross-link F-actin (Fig 1 A, B) (4,14,15). Each CH domain contains approximately 110 amino acid residues, that are arranged into four short α-helices (A, E, C, G) separated by loops (Fig 1 C) (15–17).This α-helical fold of CH1 and CH2 domains shows significant sequence and structural similarity across various proteins, particularly those involved in actin binding (spectrin, α-actinin, dystrophin, utrophin, plectin, actin-binding protein 120 (ABP-120), fibrin and cortexillin) (17).

FIGURE 1.

Schematic structure of human α-actinin-2 monomer. (A) The domain structure of human α-actinin-2 and their positions in the full-length α-actinin-2 monomer. The numbers indicate the amino acid residues. The Actin-binding domain (ABD) contains two calponin homology domains 1 and 2 (CH1 and CH2) shown in orange and blue, respectively. The residues T43, S50, S147 and T237 which are located to the CH1-CH2 interface and may be phosphorylated, are shown in magenta. The calmodulin-like domain (CaM) contains two pairs of EF-hand motifs (EF1–2 and EF3–4) shown in purple. The rod domain contains four spectrin repeats (SR1–4) shown in yellow. (B) The human α-actinin-2 resolved structure 4D1E by Ribeiro et al., 2014. The above-mentioned colors represent the same α-actinin-2 domains. The dashed black box shows the ABD. Magenta-colored spheres in the ABD depict the four phosphorylatable amino acid residues studied in the paper. The α-actinin-2 crystal structure is missing the first 33 N-terminal residues, starting with the residue number 34, and the three last C-terminal amino acid residues, ending with residue number 892. (C) The α-actinin-2 ABD in the closed conformation. CH1 domain is orange, CH2 domain blue and the loop region grey. The four α-helices are depicted with letters A, E, C, G. The blue dashed box shows the CH1-CH2 interface and the studied phosphorylation sites are shown as magenta spheres. The dark gray ribbons represent the actin-binding sites 1 and 2 (ABS1, ABS2). (D) A fraction of the resolved structure of human α-actinin-3 CH1 domain bound to F-actin. Orange represents the CH1 domains, and the different shades of grey and purple represent the actin-filament. The black dashed box indicates the contact of α-actinin-3 and actin. (E) α-actinin-3 CH1 interaction with actin monomer. Dark grey color shows the interaction between the CH1 domain and actin. The blue dashed box shows the CH1-CH2 interface.

The CH1 and CH2 domains differ from each other in their amino acid sequences and affinities. CH1 domain primarily binds F-actin, and the CH2 domain regulates CH1 binding to actin by providing steric hindrance (18, 19). α-actinin ABD has shown to contain up to three actin-binding sites (ABS), from which ABS2 and ABS3 are located on the surface. ABS1, which is positioned in the CH1-CH2 interface, is only accessible in the ABD open conformation (4, 20). Identified ABS in α-actinin-3 are R48-S57 and I153-T172 (21) and expected ABS for α-actinin-2 are Q40-N49 and L114-I138, predicted by ScanProsite (SIB Swiss Institute of Bioinformatics). Similar to other proteins containing tandem CH domains, the α-actinin-2 ABD is in the closed conformation in the absence of actin. The destabilization of the CH1-CH2 interface enhances F-actin binding (4,20). Previous MD simulations for T-plastin and fascin ABD have shown that the CH domains are capable of interdomain rotations through the loop regions connecting the CH1 and CH2 domains, as well as unfolding and extending the CH1-CH2 loop region, moving the CH domains apart to generate the actin-binding conformation (22, 23).

F-actin binds the ABD of α-actinin-2 only in the open conformation, which is achieved by phosphatidylinositol 4,5-bisphosphate (PIP2) binding the N-terminal region of CH2 domain and disrupting the hydrophobic interactions between the CaM domain and the neck region (16, 24). While the interaction between actin and α-actinin and its regulation by phospholipids like PIP2 are well described, the precise mechanisms by which sarcomeres assemble and stabilize into mature filaments in response to mechanical and neurohumoral stimuli remain unclear (18, 25, 26). Research currently focuses on the dynamic interplay of α-actinin and other proteins like titin, CapZ, and vinculin, as well as the role of mechanical force and post-translational modifications (PTMs) in regulating sarcomere assembly and stability (27, 28). Recently, it has been found that α-actinin-2 phosphorylation is upregulated in response to mechanical stimulation, increasing the phosphate groups attached to α-actinin-2 ABD. Specifically, it was found that the phosphorylation sites (T43, S50, S147, T237) developed upon mechanical stimulation, are found at the contact interface between the CH1 and CH2 domains (29).

This poses a mechanistic hypothesis in which phosphorylation regulates the binding of ABD to F-actin by altering the conformation of these domains and enhancing the opening of the CH1 and CH2 domains to facilitate more accessibility for its binding to actin (19, 30). This is supported by the observation that the CH1 and CH2 domains exhibit complementary charges at their contact surface (25).The CH1 domain, that directly interacts with F-actin, generally has a net positive charge at the interface with CH2 domain, and the CH2 domain, that acts as a negative regulator of F-actin binding through a steric clash with the actin filament upon CH1-actin complex formation, typically has a net negative charge (17, 19, 31). Phosphorylation at the CH1 domain of the ABD neutralizes such charges due to the partial negative charges added (32, 33). Second, these phosphorylation sites in the α-actinin-2 ABD have been identified in other phosphoproteomic data sets in cell types, such as human skeletal muscle cells by Højlund et al., 2009 (identified T237p) (34), human embryonic stem cells by Brill et al., 2009 (identified S147p) (35), ovarian carcinoma cells by Mertins et al., 2014 (identified S147) (36), and human T lymphocytes and cervical carcinoma cells by Mertins et al., 2013 (identified S147p) (37), suggesting that the phosphorylation of α-actinin-2, especially at the ABD, could represent generalized regulatory mechanism of cytoskeletal assembly.

Here, we studied the structural implications of α-actinin-2 phosphorylation at the ABD, by using the AlphaFold 3 server to predict conformations that shed light on how phosphorylation impacts the conformational changes of α-actinin-2. We also utilized conventional molecular dynamics (cMD) simulations, as well as umbrella sampling (USMD) experiments to explore structural changes induced by phosphomimetics. Our analyses include calculations of α-carbon root mean square deviation (C_α RMSD) of superimposed predicted structures, distances between CH1 and CH2 centers of masses to understand the alterations in the conformational dynamics, free energy potentials of the open-to-closed transition in the CH domains, the torsion angles between CH1-CH2 domains and the electrostatic interactions between α-actinin-2 variants and actin, and the Coulombic potentials of phosphorylated α-actinin-2 that mediate α-actinin-2 – actin association. These data highlight the potential role of phosphorylation in regulating α-actinin-2 conformational dynamics at the ABD to fine-tune its actin crosslinking capacity.

MATERIALS AND METHODS

AlphaFold 3 simulations

AlphaFold 3 predicted structures

Human α-actinin-2 (ACTN2_HUMAN, Uniprot entry P35609) amino acid sequence representing the actin-binding domain (ABD, residues 1–256, Fig 2 D) was utilized to represent the wild-type (WT) α-actinin-2 ABD. The residues of interest included threonine (Thr, T) at positions T43 and T237, along with serine (Ser, S) at positions S50 and S147. The protein structures for the phosphorylated ABD were modeled using the phosphorylation sites T43, S50, S147, and T237, alongside their corresponding phosphomimetic variants, which included aspartic acid (Asp, D) substitutions at T43D, S50D, S147D, and T237D, as well as alanine (Ala, A) substitutions at T43A, S50A, S147A, and T237A for control purposes. These predictions were generated using the AlphaFold 3 server (Google DeepMind & Isomorphic Labs, UK) (38), employing fifteen distinct random seeds (n=15) to ensure variability in structure predictions, and only those with a predicted template modeling (pTM) score exceeding 0.5 were selected. The structures predicted by AlphaFold 3 for α-actinin-2 WT were subsequently compared to the experimentally determined X-ray diffraction structure PDB 4D1E (4). All 15 AlphaFold 3 predicted structures for each α-actinin-2 variations were submitted to the MolProbity server (39–41) as .cif files. MolProbity server evaluated the following metrics for each model: MolProbity score, Clashscore, Ramachandran outliers, poor rotamers, Cβ deviations, bad bonds and bad angles to assess the geometry statistics and stereochemical quality compared to the experimental benchmark 4D1E PDB structure.

FIGURE 2.

Phosphorylation and pseudo-phosphorylation increase the distances between the CH1 and CH2 domain centers of mass to open the α-actinin-2 actin binding domain. (A-C) Depicts ChimeraX measured distances between calponin homology 1 (CH1) and calponin homology 2 (CH2) domains for AlphaFold3 predicted phosphorylated (A), pseudo-phosphorylated (B), and neutrally substituted (C) α-actinin-2 (ACTN2) actin-binding domain (ABD). Coloration changes from light to dark represent the phosphorylation/or substitution increase from one position to four positions. The red bar represents the WT α-actinin-2 predicted structure and the red dashed line shows the baseline for the WT structure. Two-tailed paired t-test showed statistical significance between wild-type (WT) α-actinin-2 and phosphorylated or pseudo-phosphorylated α-actinin-2 AlphaFold3 predicted structures at **p<0.01 and ***p<0.001 (α=0.001) levels, and WT α-actinin-2 and neutral substitution AlphaFold 3 predicted structures at **p<0.01 and *p<0.05 (α=0.001) levels. One-Way ANOVA showed statistical significance between groups at p<0.001 (α=0.01). (C) shows the human α-actinin-2 ABD sequence. Italic amino acid residues (residues 1–35) illustrate the N-terminal residues of the ABD that show great dispersion and were removed for simulations. CH1 domain (residues 38–142), CH2 domain (residues 151–256), and the loop region (residues 143–150) are represented in orange, blue and gray, respectively. Red bold amino acid residues depict the threonine (T) and serine (S) residues that are submitted to phosphorylation and are used for generation of phosphomimetic mutations with aspartic acid (D), as well as neutral amino acid substitution with alanine (A). Underlined regions (amino acid residues 40 – 49 and 114–138) depict the actin-binding residues. n=15 for phosphorylation, pseudo-phosphorylation and neutral amino acid substitution each. (D, E) α-actinin-2 structures show dose-dependent phosphorylation opening the α-actinin-2 ABD. Blue ribbons represent the CH2 domain, and the blue dot represents the center of mass for the CH2 domain. Orange ribbons represent the CH1 domain, and the red dot represents the center of mass for the CH1 domain. Gray coil represents the loop region, dark gray ribbons represent the actin-binding sites 1 and 2 (ABS1, ABS2), and the gray line illustrates the distance between the centers of masses. The light-yellow planes represent the torsion angle (τ°) between CH1 and CH2 domains. A higher torsion angle (WT, E) represents a more twisted or closed interdomain orientation, where CH1 and CH2 are more compactly aligned. A lower torsion angle (Phosphorylated form, F) indicates reduced twisting, allowing the domains to open up and increase the CH1-CH2 distance. Purple, green and red amino acid residues show the phosphorylated residues T43p, S50p, and T237p respectively. Two-tailed paired t-test shows statistical significance between the two illustrated α-actinin-2 AlphaFold3 predicted structures at p<0.001 (α=0.001) level. (G, H) DelPhi Force-calculated electrostatic forces illustrate the effects of phosphomimetic and alanine substitutions on the structure and actin-binding properties of α-actinin-2 ABD variants. (G) Comparison of phosphomimetic α-actinin-2 ABD variants (S147D, S50D, T237D, T43D) with the wild-type (WT) ABD. (H) Comparison of alanine-substituted α-actinin-2 ABD variants (S147A, S50A, T237A, T43A) with WT ABD. Scatter plot shows the relationship between the CH1–CH2 domain center distance (Å) and torsion angle (°) derived from ChimeraX measurements of AlphaFold3-predicted structures. Color scale represents electrostatic contribution to binding free energy ( ΔH_elec, kcal/mol), with warmer colors indicating less favorable electrostatic interactions and cooler colors indicating more favorable electrostatic interactions. ΔH_elec reflects the electrostatic contribution to interaction energy calculated by Delphi Force and does not represent binding affinity or total binding free energy, which also includes non-electrostatic and entropic contributions.

ChimeraX structural calculations

Crystallographic information files (.cif) predicted by AlphaFold 3 were utilized for simulations within the UCSF ChimeraX software (University of California San Francisco, UCSF, CA) (42). In the simulations of the α-actinin-2 ABD, the N-terminal tails exhibiting significant conformational variability (amino acid residues 1–35) were excluded. Each structure, whether phosphorylated, pseudo-phosphorylated, or neutrally substituted, was aligned with the WT α-actinin-2 ABD using the matchmaker structure analysis tool, which employs pairwise sequence alignment. C_α RMSD was computed to evaluate the deviation among the superimposed structures, utilizing matchmaker alignments for each atomic position of the amino acid residues, alongside the median C_α RMSD for both pruned atom pairs and all atom pairs. The centers of mass and geometric centers for the CH1 domain (residues 36 – 142) and CH2 domain (residues 151–256) were determined using the” define centroid mass weighting” and” define centroid” commands, respectively. The distance between the centroids of each modified α-actinin-2 predicted structure was measured using the structure measurements tool. The Coulombic potential for the α-actinin-2 ABD, CH1 domain, CH2 domain, actin-binding residues 40–49 (ABS1), and actin-binding residues 114–138 (ABS2) was calculated individually through the Coulombic command. The Coulombic potentials reported correspond to the absolute electrostatic energy of each individual configuration, calculated independently for each domain or residue set, without reference to other conformations. The data was statistically analyzed and graphed in GraphPad Prism software (GraphPad Software, Boston, MA), based on the average, standard deviation, and number of replicates for each structure. One-Way ANOVA statistical analysis was used for statistical comparisons among all independent structures and the two-tailed paired t-test for statistical comparisons between two structures (WT α-actinin-2 ABD structure versus modified structure). A p-value of less than 0.05 was deemed statistically significant. The torsion angle for each structural model was calculated in ChimeraX using the torsion command, defining the dihedral angle between the four sets of residues: 49–51, 38–40, 245–247, and 254–256.

Electrostatic Force calculations

AlphaFold3 generated structural models of α-actinin-2 WT, phosphorylated, pseudophosphorylated, and alanine-substituted variants .cif files were uploaded to the DelPhiPKa web server (43, 44) for protonation and pKa calculations under physiological conditions (0.15 M ionic strength, pH 7.0). The protonated structures were converted to .pqr format and subsequently submitted to the DelPhi Force web server (45) to compute electrostatic interaction forces. For these calculations, the reference (receptor) α-actinin-2 molecule was paired with an actin monomer (ligand) extracted from a pre-processed α-actinin–actin complex .pqr file. Interactions were evaluated using the parameters: protein dielectric (ε_prot) 40; solvent dielectric (ε_solv) 80; physiological ionic strength 0.15 M; temperature 298 K; default grid resolution and default focusing. The resulting electrostatic contribution to binding free energy (change in electrostatic energy, ΔH_elec ) for each variant (n=15) were used as raw output from the server in kilocalories per mole (kcal/mol). Delphi Force calculations report the electrostatic contribution to interaction energy ( ΔH_elec), which represents only one component of binding thermodynamics and should not be interpreted as binding affinity or total free energy. Structural parameters from ChimeraX, including CH1-CH2 interdomain distances and torsion angles, together with the calculated ΔH_elec values were combined into a single dataframe in RStudio (v4.5.1). Scatter plots of CH1-CH2 distance versus torsion angle were generated using ggplot2, with points colored according to ΔH_elec values in kcal/mol (blue = low, red = high) and shapes indicating variant identity. Legends and axis labels were included to facilitate visualization of the relationship between ABD conformation and electrostatic interaction energies.

Molecular dynamics simulations

Model Construction

We generated two all-atom atomic models of the ABD of human α-actinin-2 (henceforth, actinin). The first models the actinin ABD in an actin-unbound (or ‘closed’) conformation and is based on the 3.5 Å resolution X-ray crystal structure of human α-actinin-2 (PDB: 4D1E) (4). We extracted coordinates of residues 34–258 from the X-ray structure and then modeled in 4 disordered N-terminal residues (LLD) using Modeller (46). The second models the actinin ABD in an actin-bound (or ‘open’) conformation and is based on the 2.6 Å resolution cryo-EM structure of human plastin 3 in complex with chicken skeletal actin (PDB: 7R94) (22). We used the Swiss-Model server (47) to construct a homology model using the sequence of human α-actinin-2 (residues 34–258) and plastin 3 coordinates as a template. Due to the conservation of this actin binding motif, our model of the actin-bound conformation was compatible with a human actin structure (Fig 1 D). Next, we generated 3 phosphomimetic structures (T43D, S147D, T237D) for both the actin-bound and actin-free models using the ‘Rotamers’ tool within UCSF Chimera (42). This resulted in 8 all-atom structures: four unique sequences (WT, T43D, S147D, T237D) in two conformations (actin-bound and actin-free).

Simulation Preparation

The eight models were prepared for all-atom MD simulations using the AMBER20 simulation program (48) following our standard protocol (49). Atomic systems were prepared using the ff14SB protein force field (50) and Li and Merz (51) ion force field. The proteins were solvated in truncated octahedral boxes with a maximum diameter of ~85 Å, filled with ~16000 TIP3P (46) waters and ~20 KCl ion pairs (~70 mM KCl). Then the systems were energy-minimized in three stages: hydrogens were minimized for 1000 steps, solvent atoms were minimized for 1000 steps, and finally all atoms were minimized for 8000 steps in the presence of 25 kcal mol ⁻¹ Å⁻² restraints on protein heavy backbone atoms. The systems were then heated to 310 K over 150 ps in the NVT ensemble (using the Langevin thermostat and collision frequency of 2 ps⁻¹). Finally, the systems were density-equilibrated over 5.2 ns in the NPT ensemble (using the Langevin thermostat and Monte Carlo barostat). During the heating and equilibration procedure, restraints on protein heavy backbone atoms were decreased from 25 to 0 kcal mol ⁻¹ Å⁻².

Relationship between AlphaFold-generated structures and equilibrated MD conformations

The AlphaFold-generated ABD models begin in a CH1-CH2 separation of approximately 26 Å, which lies slightly outside the lowest-energy basin identified in the umbrella-sampling free-energy landscape. During the equilibration phase in explicit solvent, these starting structures undergo modest relaxation of the CH-domain geometry toward the dominant free-energy minima sampled in the MD simulations. Thus, the AlphaFold-predicted models are used as physically reasonable starting structures rather than as solution-phase free-energy minima.

Production Simulations

Conventional MD (cMD) production simulations were carried out in the NVT ensemble using the Bussi thermostat at 310 K using a 2 fs timestep and a 9 Å nonbonded cutoff. Simulations were performed in triplicate for 1000 ns. Coordinates were saved every 10 picoseconds for analysis. This yielded 24 μseconds of net sampling across all cMD trajectories.

Production Simulations Umbrella Sampling MD (USMD)

We additionally performed umbrella sampling molecular dynamics (USMD) simulations to estimate the free energy landscape along the open-to-closed inter-CH distance reaction coordinate. Simulations initiated in the open/actin-bound conformation adopted closed/actin-free conformations in our cMD simulations. Therefore, initial coordinates for the USMD simulations were extracted from our cMD simulations and umbrella windows along the reaction coordinate were prepared. We identified a smooth path spanning inter-CH distances of 26.7 Å to 33 Å that was sampled by all four genotypes during the cMD simulations and selected 63 conformations per genotype along the path (0.1 Å spacing between window centers). Our reaction coordinate was the center of mass distance between residues 42–128 (CH1) and residues 159–259 (CH2). The USMD simulations were prepared identically to the cMD simulations with the addition of 10 kcal mol⁻¹ Å⁻² umbrella restraints, and single-run USMD production simulations were performed with 20 ns of sampling in each umbrella window. Inter-CH distances were recorded every 0.1 ps and free energies along the reaction coordinate were determined via the WHAM code of Alan Grossfield (52).

Simulation Visualization and Analysis

C_α RMSDs, root mean square fluctuations (C_α RMSFs), interatomic distances, CH1/CH2 domain distances, CH1/CH2 domain torsions, and interresidue contacts were calculated with cpptraj (53). Two residues were considered in contact if at least one pair of heavy atoms between the residues were within 5 Å of one another. Contact frequencies were calculated as the percentage of frames in the production simulations for which residues were in contact. Coulomb energies were calculated as the absolute electrostatic energy of each individual configuration using the AMBER20 energy decomposition tools, without reference to other conformations. USMD potential of mean mass (PMFs) was constructed using WHAM (52). Protein images were created with UCSF Chimera (42).

RESULTS

AlphaFold 3 predicted α-actinin-2 actin-binding domain structure

In this study, the AlphaFold 3 algorithm was employed to generate structural predictions of the α-actinin-2 ABD for WT, phosphorylated, pseudo-phosphorylated, and neutrally substituted variants, utilizing different random seeds (n=15). The AlphaFold 3 server produced the WT α-actinin-2 ABD with a notably high predicted template modeling (pTM) score of 0.86 (Table S14), indicating a robust level of confidence in the prediction. The region encompassing amino acid residues 35–256, which contains the calponin CH1 and CH2 domains, exhibited the highest predicted local distance difference test (plDDT) value exceeding 90. Conversely, the N-terminal tail displayed the lowest plDDT value, with scores below 50 for residues 1–30 and between 50 and 70 for residues 31–34. The predictions from AlphaFold 3 corroborated the significant conformational variability of the N-terminal tail (4), consistent with expectations for disordered regions, which were subsequently excluded from the AlphaFold 3 predicted structures to reduce computational inaccuracies.

To date, the only WT human α-actinin-2 ABD that has been resolved through X-ray diffraction is the PDB 4D1E structure reported by Ribeiro et al. in 2014, which exhibits a closed conformation (4). The 4D1E α-actinin-2 structure served as a control to assess the accuracy of the AlphaFold 3 predicted WT α-actinin-2 ABD structure. The mean C_α RMSD value for the superimposed ABD of 4D1E and the AlphaFold 3 WT α-actinin-2 was found to be 0.66 ± 0.09, with comparable C_α RMSD values observed in both the CH1 and CH2 domains (Table S16), indicating minimal variations in the backbone atoms. Additionally, the distances measured between the centers of mass of CH1 and CH2, along with geometric distances, did not show statistical significance (p=0.29 and p=0.06, respectively) between the two structures, suggesting that AlphaFold 3 accurately predicts the α-actinin-2 ABD WT in a closed conformation.

The neutrally substituted variants were produced with a high degree of certainty, exhibiting pTM scores of 0.85 and 0.86 (Table S14), and a plDDT exceeding 90 in the range of 35–256. Similarly, the phosphomimetic variants were predicted with substantial confidence, showing pTM scores ranging from 0.85 to 0.86 (Table S14), along with a plDDT above 90 in the same region.

The models of phosphorylated α-actinin-2 exhibited lower accuracy, with pTM scores between 0.62 and 0.86 (Table S14), and plDDT values from 70 to over 90 for the ABD. Notably, all α-actinin-2 variants that included phosphorylation at the T43 residue recorded the lowest pTM scores of 0.62 to 0.69. In contrast, the remaining α-actinin-2 models, which only featured phosphorylation at S50, S147, and T237, demonstrated high prediction accuracy with pTM values of 0.84 to 0.86 and plDDT scores exceeding 90 for residues 35 to 256. AlphaFold 3 has provided reliable predictions for covalently modified structures, such as phosphorylated proteins, achieving a success rate of approximately 50–70%, compared to nearly 90% for unmodified protein monomers (38). This indicates that structural alterations caused by phosphorylation, particularly when multiple residues are modified, may significantly affect structure predictions and warrant careful interpretation (54).

The AlphaFold 3 predicted α-actinin-2 structures were further evaluated by MolProbity server to assess their stereochemical quality in comparison to the 4D1E PDB structure. The MolProbity results (Fig S1; Table S1) demonstrated that the AlphaFold3 models for the α-actinin-2 ABD, including phosphorylated, pseudo-phosphorylated, and alanine-substituted variants, maintain consistently high structural quality across all levels of substitution (mono-, di-, and tri/tetrasubstituted). In each case, the MolProbity scores remain close to the benchmark experimental 4D1E ABD structure, indicating that the predicted models exhibit low steric clashes and well-defined local geometry. The phosphorylated and pseudo-phosphorylated variants show particularly favorable percentile ranks, suggesting that the introduction of these modifications does not significantly disrupt the backbone geometry or side-chain packing. Similarly, the alanine-substituted variants, which serve as non-phosphorylatable controls, exhibit MolProbity and Clashscore values comparable to both the WT and modified structures, reinforcing the overall robustness of AlphaFold3 predictions. Across the substitution series, the transition from mono- to di- and tri/tetrasubstituted variants does not result in a marked decline in structural quality metrics such as Ramachandran outliers, poor rotamers, or bad bonds and angles. This consistency suggests that AlphaFold3 can reliably accommodate multiple site modifications without introducing major structural artifacts. Collectively, these MolProbity results support the conclusion that AlphaFold3 produces geometrically sound and stereochemically accurate models for both WT and modified α-actinin-2 ABD variants, making them suitable for interpreting the potential structural and functional effects of phosphorylation and substitution events further evaluated in this study.

Distance between the calponin homology domains

The structures of α-actinin-2 ABD generated by AlphaFold 3 were analyzed using UCSF ChimeraX software to evaluate the conformational alterations induced by phosphorylation on the open-closed state of the ABD, specifically by measuring the distance between the centers of mass of the CH1 and CH2 domains. The phosphorylation of ABD led to a dose-dependent destabilization of the closed-state conformation (Fig 2 A, E–F). The tri- and tetra-phosphorylated variants exhibited the greatest distances between the CH1 and CH2 centers of mass, measuring 27.03 ± 1.98 Å for α-actinin-2 T43p S147p T237p; 26.87 ± 1.30 Å for α-actinin-2 T43p S50p T237p; and 26.27 ± 0.09 Å for α-actinin-2 T43p S50p S147p T237p, all showing a statistically significant difference (p < 0.001) when compared to the AlphaFold 3 predicted WT α-actinin-2 structure and the PDB 4D1E structure (Fig 2 A, E–F; Table S2). Similarly, the di-phosphorylated and mono-phosphorylated α-actinin-2 ABD also demonstrated an increase in the distance between the CH1 and CH2 centers of mass, revealing a significant difference from the WT α-actinin-2 and suggesting a dose-dependency between number of phosphorylation sites and conformational distancing between the CH1 and CH2 domains.

The formation of a phospho-ester bond with S or T residue during phosphorylation leads to an increase in mass of 80 Da at the specific residue within the protein, thereby altering the mass distribution and displacing the centers of mass of the phosphorylated domain (55). To accommodate this mass increase, the geometric centers of the CH1 and CH2 domains were evaluated to consider any dispersion in the center of mass calculations. The geometric distance data (Fig S2 A; Table S3) corroborated the findings and validated the conformational alterations induced by phosphorylation. Notably, similarly to the low pTM values, the variants featuring T43 phosphorylation exhibited the greatest distances between the CH domains, demonstrating the highest standard deviation values and variations, including a fully opened ABD conformation at 31.89 Å in the α-actinin-2 T43p S147p T237p model and 31.08 Å in the α-actinin-2 T43p S50p model. It remains unclear whether this phenomenon is attributable to a greater degree of inaccuracy in the AlphaFold 3 predictions, which correlate with the lower pTM values for these structures, or if phosphorylation of the T43 residue influences the opening of the α-actinin-2 ABD through conformational effects.

Because phosphorylation is more precisely studied in vitro by phosphomimetic substitutions (i.e., via S/T to D/E single point mutations) via molecular biology techniques, the latter substitutions were studied. Phosphomimetic variants exhibit a comparable dose-dependent response to the predictions made for phosphorylated α-actinin-2. The tri- and tetra-pseudo-phosphorylated α-actinin-2 variants display the greatest distance values between the centers of mass of CH1 and CH2, with the T43D S147D T237D analog presenting the most expanded configuration at a CH1-CH2 distance of 26.41±0.37 Å. This is followed closely by T43D S50D T237D at 26.29 ± 0.29 Å and T43D S50D S147D T237D at 26.30 ± 0.31 Å (p < 0.001) (Fig 2 B; Table S2). The mono- and di-pseudo-phosphorylated models exhibit a slightly lesser degree of opening, yet they demonstrate statistical significance when compared to the α-actinin-2 WT ABD (Fig 2 B). Notably, when juxtaposed with their phosphorylated equivalents, modest differences are seen (e.g. Figs. 2 A vs 2 B and 3 A vs 3 B), and the overall shift in distances remains consistent with an opening induced by phosphorylation. The residual discrepancies may stem from the fact that Asp/Glu substitutions do not fully mimic the charge, geometry, or hydrogen-bonding features of a true phosphate group. This assertion is further corroborated by geometric distance measurements, which confirm that dose-dependent pseudo-phosphorylation enhances the open conformation of the α-actinin-2 ABD (Fig S2 B). Concurrently, phosphomimetic α-actinin-2 models reveal a similar trend, with the involvement of the T43 residue contributing to an increased distance between the CH1 and CH2 centers of mass and facilitating the opening of the ABD conformation, thereby suggesting a stabilizing role of T43 at the CH1-CH2 interface. Thereby indicating that phosphorylation likely promotes opening of the α-actinin-2 ABD, albeit with modest differences between true phosphorylated and phosphomimetic forms.

FIGURE 3.

Phosphorylation and pseudo-phosphorylation increase the C_α RMSD values in superimposed structures of α-actinin-2 actin binding domain in a dose-dependent manner. ChimeraX measured distances between calponin homology 1 (CH1) and calponin homology 2 (CH2) domains for Alphafold3 predicted phosphorylated (A), pseudo-phosphorylated (B), and neutrally substituted (C) α-actinin-2 actin-binding domain (ABD). (1–3) exhibit the C_α RMSD values for each amino acid residue in the ABD separately. (1) show the mono-phosphorylated, mono-phosphomimetic and mono-substituted α-actinin-2 ABD for residues T43, S50, S147 and T237 separately. (2) show the bi-phosphorylated, bi-phosphomimetic and bi-substituted α-actinin-2 ABD, using different combinations of the forementioned four residues. (3) show the tri- and tetra-phosphorylated/ phosphomimetic and substituted α-actinin-2 ABD, where different combinations of the four previously mentioned amino acid residues were used. Solid lines represent the mean C_α RMSD values for each amino acid residue, and the dashed line represents the error bars between each measured prediction. (4) depicts the mean C_α RMSD value for all the pruned amino acid residues. The structures show the superimposed AlphaFold3 α-actinin-2 WT to modified α-actinin-2 where C_α RMSD values are rendered by color, with green coloration referring to a high degree and orange coloration referring to a moderate degree of structural similarity between the α-actinin-2 WT and the modification. Purple coloration refers to a significant difference between the structures. Red dashed boxes show the amino acid residues which have been phosphorylated or substituted by aspartic acid (D) or alanine (A). One-Way ANOVA showed statistical significance between groups at p<0.001 (α=0.01). n=15 for phosphorylation, pseudo-phosphorylation and neutral amino acid substitution each.

The neutrally substituted negative controls (i.e., S/T to A) exhibit a dose-dependent pattern, with α-actinin-2 T43 S50A T237A, α-actinin-2 T43A S147A T237A, and α-actinin-2 T43A S50A S147A T237A displaying the greatest distance measurements between the CH1 and CH2 domains (25.95 ± 0.11 Å, 25.94 ± 0.07 Å, and 25.93 ± 0.07 Å, respectively), although with limited statistical significance (p ≤ 0.05). In contrast, the remaining tri-, di-, and mono-substituted variants did not show any statistical significance when compared to the α-actinin-2 WT models, suggesting that the incorporation of the negative side chain from aspartic acid (D) or phosphate, alters the ABD conformation and facilitates the CH1-CH2 opening mechanism (Fig 2 C; Table S2).

Average distance between the α-carbons of superimposed α-actinin-2 predictions

In a manner analogous to the distance measurements between the centers of mass of CH1 and CH2, the average C_α RMSD values for all pruned atoms were computed for the entire ABD, as well as for the individual CH1 and CH2 domains, to evaluate the structural alterations resulting from each modification in relation to the α-actinin-2 WT. Furthermore, C_α RMSD values were determined for each pruned atom individually to identify regions exhibiting the greatest deviations, including the ABS 40–49 and ABS 114–138 segments, to ascertain whether these conformational changes influence the binding affinity for F-actin.

In relation to the distances between the centers of mass of CH1 and CH2, the increase in the C_α RMSD value exhibited a dose-dependent response to the modifications, with phosphorylation demonstrating the most significant deviations from the atomic positions when compared to the α-actinin-2 WT, particularly in structures containing the T43 residue (Fig 3 A). The most pronounced structural variation in the complete ABD was observed for α-actinin-2 T43p S50p T237p, with a C_α RMSD of 3.12 ± 3.08 Å; α-actinin-2 T43p S147p T237p at 2.79 ± 2.98 Å; and α-actinin-2 T43p S50p at 2.29 ± 2.92 Å (Table S4), which correlates with the distance measurements for the CH1-CH2 centers of mass. These findings indicate a notable divergence from the α-actinin-2 WT, potentially reflecting a more open conformation. The variants α-actinin-2 T43p and α-actinin-2 T43p S50p S147p T237p exhibit minor differences, while the remaining phosphorylated variants show a high degree of structural similarity to the α-actinin-2 WT. When evaluating the C_α RMSD values for the CH1 and CH2 domains independently, a greater alignment is observed, particularly in the CH1 domain, which has average C_α RMSD values below 0.2 Å. The CH2 domain; however, displays slightly more variability.

The structural analysis of ABS 40–49 reveals nearly identical characteristics, with mean C_α RMSD values between 0.03 and 0.11 Å (Table S7), while the ABS 114–138 region exhibits slight variations, with mean C_α RMSD values ranging from 0.13 to 0.19 Å (Table S8). This suggests that the observed conformational changes may not significantly impact F-actin binding. A detailed comparison of C_α RMSD values for individual amino acid residues in the aligned ABD, as well as in the aligned CH1 and CH2 domains (Fig S3; Table S5 and S6, respectively), indicates that the CH1 domain maintains alignment. On the other hand, the CH2 domain shows considerable fluctuation with C_α RMSD values reaching up to 10 Å for tri-phosphorylated variants. This observation implies a potential opening of the ABD conformation due to the untangling of the loop region, which is further supported by the highest C_α RMSD values recorded for the loop region within the CH1 and CH2 domains (Fig S4), corresponding with the distance measurements between the centers of mass of CH1 and CH2.

The phosphomimetic variants of α-actinin-2 serve as positive controls, demonstrating a dose-dependent response similarly to that of the phosphorylated α-actinin-2 models, although with reduced mean C_α RMSD values (Fig 3 B). The variants α-actinin-2 T43D S147D T237D, α-actinin-2 T43D S50D S147D T237D, and α-actinin-2 T43D T237D exhibited the highest mean C_α RMSD values for the ABD, measuring 0.66 ± 0.71 Å, 0.58 ± 0.67 Å, and 0.44 ± 0.35 Å, respectively. This suggests that the structures of pseudo-phosphorylated α-actinin-2 are very similar to those of WT α-actinin-2 (Fig 3 B; Fig S3). In comparison to the phosphorylated variants, most models did not show statistically significant differences, with the exceptions of α-actinin-2 T43p S50p T237p (p < 0.01), α-actinin-2 T43p S147p T237p (p < 0.05), and α-actinin-2 T43p S50p (p < 0.05). Several AlphaFold 3 predictions for these phosphorylated structures indicated an open conformation, with mean C_α RMSD values exceeding 6 Å (Table S4). Similar to the phosphorylated models, the CH1 and CH2 domains were closely aligned with the WT, exhibiting C_α RMSD values between 0.12 – 0.21 Å for CH1 and 0.14 – 0.28 Å for CH2. These results reveal greater variability in the loop regions when aligning these domains separately (Fig S5), which may suggest a potential opening of the ABD conformation. In comparison to phosphorylation, the ABS 40–49 and ABS 114–138 regions are nearly identical to WT α-actinin-2, with mean C_α RMSD values of 0.02 – 0.06 Å and 0.07 – 0.10 Å, respectively (Table S7 and S8, respectively).

The neutrally substituted α-actinin-2 models, utilized as negative controls, demonstrated a significant resemblance to the WT α-actinin-2, with all variants exhibiting comparable mean C_α RMSD values for the entire ABD, ranging from 0.13 to 0.19 Å (Fig 3 C). Furthermore, the mean C_α RMSD values for the CH1 and CH2 domains were lower, falling between 0.10 – 0.13 Å and 0.11 – 0.14 Å, respectively, while the ABS 40–49 and ABS 114–138 regions displayed nearly identical structures to the WT α-actinin-2, with values of 0.02 – 0.05 Å and 0.07 – 0.08 Å (Table S4). All corresponding phosphorylated and neutrally substituted α-actinin-2 models exhibited statistically significant differences in their full ABD, primarily at p < 0.01 and < 0.001. Additionally, all analogous phosphomimetic and neutrally substituted α-actinin-2 variants showed statistical significance in the ABD, predominantly at p < 0.01 and < 0.001, indicating structural variations within the ABD.

Coulombic potential

The net Coulombic potential of the complete ABD, as well as the CH1 and CH2 domains individually, along with the ABS 40–49 and ABS 114–138 segments, was assessed using ChimeraX software for all AlphaFold 3 predictions and were subsequently compared. Phosphorylation of α-actinin-2 enhances the negative net Coulombic potential of the ABD in a dose-dependent manner (Fig 4 A). The average Coulombic potential for the WT α-actinin-2 is −0.73 ± 0.02 kcal/(mol·e), which increases to −2.33 ± 0.12 kcal/(mol·e) for the α-actinin-2 T43p S50p S147p T237p variant; −2.22 ± 0.14 kcal/(mol·e) for α-actinin-2 T43p S147p T237p; and −2.20 ± 0.12 kcal/(mol·e) for α-actinin-2 T43p S50p T237p. Even the mono-phosphorylated variants exhibit a net Coulombic potential exceeding −1.19 kcal/(mol·e) for the ABD, with all phosphorylated forms demonstrating a significant statistical difference (p < 0.001) when compared to the predictions for α-actinin-2 WT (Table S9). Visualization of the Coulombic map reveals an augmented negative electrostatic surface potential at the CH1-CH2 interface for both CH1 and CH2 domains: from 0.88 ± 0.06 kcal/(mol·e) for α-actinin-2 WT to −0.84 ± 0.04 kcal/(mol·e) for the tetra-phosphorylated model, and from −2.41 ± 0.03 kcal/(mol·e) for α-actinin-2 WT to −3.32 ± 0.13 kcal/(mol·e) for tetra-phosphorylated α-actinin-2, respectively (Table S10, S11). The mean Coulombic potential for the ABS 40–49 region exhibited negative values for all α-actinin-2 variants containing the T43p residue, ranging from 2.32 ± 0.02 kcal/(mol·e) in α-actinin-2 WT to −0.82 ± 0.09 kcal/(mol·e) in α-actinin-2 T43p S147p T237p, while the Coulombic potential in the ABS 114–138 remained unchanged (Table S12). The alteration in surface electrostatic potential within the ABS 40–49 region could influence the binding of F-actin to the CH1 domain (Table S13).

FIGURE 4.

Phosphorylation and pseudo-phosphorylation of all the studied residues increase the net negative electrostatic potential at the CH2 domain while causing a net positive to net negative transition at the CH1 domain. (A-C) ChimeraX measured Coulombic potential separately for actin-binding domain (ABD, residues 36–256), calponin homology domain 1 (CH1, residues 38–142), calponin homology domain 2 (CH2, residues 151–256), actin-binding site 1 (ABS1, residues 40–49), and actin-binding site 2 (ABS2, residues 114–138) in AlphaFold 3 predicted α-actinin-2 phosphorylated (A), pseudo-phosphorylated (B) and neutrally substituted (C) structures, in comparison to the WT α-actinin-2. Indication by double gradient colormap of GraphPad with the red color refers to the most negative Coulombic potential value at −3.5, and the blue color refers to the most positive Coulombic potential value at +2.5. Coulombic potential is shown in kcal/(mol·e). (D, E) show the Coulombic electrostatic potential (ESP) maps of α-actinin-2 WT and phosphorylated, pseudo-phosphorylated and neutrally substituted versions. (D) Dashed black rectangles indicate the interface between the CH1 and CH2 domains, with the CH1 domain being located to the left and the CH2 domain located to the right. (E) The middle figure represents the surface ESP of the ABD from the other side of the molecule in comparison to the D figure, showing CH2 domain up top and the CH1 domain on the bottom. The molecule was cut between the two CH domains and moved apart, then rotated 90° left (on the left side) and right (on the right side) showing the complementarity of the interacting surfaces. n=15 for phosphorylation, pseudo-phosphorylation and neutral amino acid substitution each. ESP red coloration indicates the highest negative potential at the value of −10 kcal/(mol·e) and blue coloration indicates the highest positive potential at the value of +10 kcal/(mol·e). One-Way ANOVA showed statistical significance between groups at p<0.001 (α=0.01). n=15 for phosphorylation, pseudo-phosphorylation and neutral amino acid substitution each.

The phosphomimetic analogues exhibit a comparable dose-dependent response, enhancing the negative Coulombic potential in the α-actinin-2 ABD, albeit to a lesser extent than phosphorylation (Fig 4 B). This is attributed to the phosphate group contributing a double negative charge, while aspartic acid contributes only a single negative charge. Statistical analysis reveals significant differences among all phosphorylated and phosphomimetic analogues, with p-values ranging from p<0.001 to p<0.05. The measured net Coulombic potential for the ABD is −1.93 ± 0.06 kcal/(mol·e) for tetra-pseudo-phosphorylated α-actinin-2 and −1.72 ± 0.11 kcal/(mol·e) for α-actinin-2 T43D S147D T237D. All phosphomimetic variants demonstrate a significant statistical difference (p < 0.001) when compared to the α-actinin-2 WT, similar to the phosphorylated analogues (Table S9). Furthermore, pseudo-phosphorylation enhances the negative surface electrostatic potential in the CH1 and CH2 domains individually and increases the negative charge at the CH1-CH2 interface. Additionally, comparably to phosphorylation, pseudo-phosphorylation induces a slight alteration in the Coulombic potential within the ABS 40–49 region in the tri- and tetra-phosphomimetic variants, while no changes are observed in the ABS 114–138 region (Table S12, S13).

In a corresponding manner to the calculations of the center of mass distance between CH1 and CH2, as well as the C_α RMSD assessments, the variants T43p and T43D exhibit an elevation in the average negative Coulombic potential on the surface, which results in the repulsion of the CH1-CH2 domains and facilitates the opening of the ABD conformation. Furthermore, the electronegativity of the CH2 domain interacting surface is enhanced in a dose-dependent manner through phosphorylation and pseudo-phosphorylation, while the interacting surface of the CH1 domain transitions from a net positive to a net negative charge (Fig 4 D, E).

The models with neutral substitutions representing negative controls exhibit no statistically significant differences in the net Coulombic potential in the ABD, CH1, CH2, ABS 40–49, or ABS 114–138 when compared to the WT α-actinin-2, as the uncharged S and T residues are replaced by a neutral A residue (Fig 4 C). In contrast, a significant statistical difference is observed when these models are compared to their phosphorylated and pseudo-phosphorylated counterparts, highlighting the substantial impact that phosphorylation of the specific four amino acid residues has on the Coulombic potential of the entire ABD, CH domains, and actin-binding sites.

Conformational and electrostatic determinants of actin binding

The balance between open and closed conformations of α-actinin-2 is influenced by electrostatic, solvation, and hydrophobic interactions. Our MD simulations, performed at physiological KCl concentration, better encapsulate these effects than static structural models. To evaluate the contribution of electrostatic interactions to actin binding under physiological conditions, we analyzed the AlphaFold3-generated α-actinin-2 structures using DelPhiForce after protonation with DelPhiPka at 0.15 M ionic strength and pH 7.0.The calculated changes in the electrostatic interaction energy ( ΔH_elec, kcal/mol), shown in Table S17, allow direct comparison of the electrostatic contribution to actin binding among different conformational states. Structural analysis of the α-actinin-2 ABD revealed that phosphomimetic substitutions (S147D, S50D, T237D, and T43D) induced a clear conformational shift relative to the WT protein (Fig 2 G). These variants displayed increased CH1-CH2 interdomain distances and lower torsion angles, indicative of a more open ABD configuration. The conformational change correlated with lower (more favorable) ΔH_elec values, indicating enhanced electrostatic stabilization in phosphomimetic variants. In contrast, alanine substitutions at the same residues (S147A, S50A, T237A, and T43A) showed CH1-CH2 distances, torsion angles, and ΔH_elec values comparable to WT (Fig 2 H), indicating minimal structural and electrostatic perturbation. The torsion angle between the CH1 and CH2 domains decreased in the phosphorylated and pseudophosphorylated α-actinin-2 ABD (Fig 2 F) compared with the WT (Fig 2 E), reflecting a reduction in interdomain twist. This decrease corresponds to a more open conformation with increased CH1-CH2 separation, consistent with exposure of the actin-binding surface and more favorable electrostatic interactions observed for phosphorylated and phosphomimetic variants. Notably, the AlphaFold-predicted ABD structures begin in a slightly more compact (closed) CH1-CH2 conformation (~26 Å) than the MD free-energy landscape (Fig. 5) and relax toward these more open equilibrium states during MD equilibration. Similarly, phosphorylated α-actinin-2 ABDs show a marked shift toward increased CH1-CH2 separation and decreased torsion angles relative to the WT, reflecting a transition to a more open conformation with more favorable electrostatic interactions. The di- and tri-substituted phosphorylated (Fig S7 D, G) and phosphomimetic (Fig S7 E, H) variants further accentuate this effect, exhibiting the greatest interdomain distances and lowest torsion angles observed. These data indicate a cumulative impact of multiple substitutions on ABD opening, suggesting cooperative modulation of interdomain flexibility and actin-binding affinity through multi-site phosphorylation. This supports a model in which phosphorylation dynamically regulates α-actinin-2 actin-binding through modulation of ABD interdomain flexibility. This analysis revealed that the open conformation of the α-actinin-2 ABD exhibits a higher predicted affinity toward actin than the closed conformation, consistent with the increased exposure of key charged residues at the binding interface. These electrostatic findings align with our MD simulations, which also suggest that phosphorylation-dependent shifts in conformational equilibrium modulate actin-binding propensity. Analysis of the electrostatic component of the binding energy ( ΔH_elec) using Delphi Force further indicates that phosphorylation or phosphomimetic substitutions increase the electrostatic contribution to domain separation, whereas non-phosphorylatable alanine substitutions maintain electrostatic interactions similar to wild-type. These observations highlight that electrostatic enthalpy is a major determinant of the phosphorylation-induced conformational changes, without implying changes in the total free energy, which also includes additional contributions such as configurational entropy or dispersion.

FIGURE 5.

MD simulations show that all the phosphomimetic mutations increased the spread in CH1-CH2 torsional and distance-dependent conformations. We calculated the CH1-CH2 interdomain distance (x-axes) and relative rotation (y-axes) for the WT and phosphomimetic MD simulations. Data from all 6 simulations (3x actin-bound-like and 3x actin-free-like) were combined. These data are presented as 2D landscapes where free energies were calculated from normalized population histograms according to the Boltzmann relation. The data from all WT (A), T43D (B), S147D (C), and T237D (D) cMD simulations are shown here, and results for the separate actin-bound-like and actin-unbound-like simulations are shown in Figure S8 The histogram bins are colored according to the apparent free energy within each basin and all FE values are expressed relative to the maximally populated bin among all simulated sequences. Cyan circles denote the conformation of the initial actin-bound-like model and pink squares denote the conformation of the initial actin-free-like model. USMD-derived FE profiles of the open-closed transition were calculated using the WHAM method for all sequences and the T43D (E, pink), S147D (F, yellow), and T237D (G, green) profiles were compared against WT (black).

Phosphomimetic residues modulated the CH1-CH2 torsional orientation and inter-domain distance

To determine the effects of the T43D, S147D, and T237D phosphomimetic mutations on the dynamics of the ABD we next performed conventional molecular dynamics (cMD) simulations. We simulated each of the four sequences (WT, T43D, S147D, and T237D) in two conformational states: an actin-bound-like conformation and an actin-free-like conformation (Table 1, Supplemental movies SM1 and SM2) and triplicate simulations were performed in each conformational state for each sequence for 1 microsecond (24,000 nanoseconds of net sampling). We measured two geometric properties of the ABD: the distance between the two CH domains and the degree of twisting between the two CH domains. We calculated the distance as the center of mass between CH1 (residues 38–124) and CH2 (residues 155–255). We calculated the twist as a center of mass dihedral angle between four groups of residues: 49–51, 38–40, 245–247, and 254–256. In the actin-free-like conformation, the two CH domains are separated by ~28 Å with a twist of ~18°. In the actin-bound-like conformation, the two CH domains are separated by ~31 Å with a twist of −0.6°. Figure 5A–D shows the sampling of these metrics for all WT and phosphomimetic systems (data from the actin-free-like and actin-bound-like simulations are combined). Across all simulations, the average WT interdomain distance was 28.7 Å. This distance was larger for T43D, S147D, and T237D (29.8 Å, 29.6 Å, and 29.2 Å, respectively). The CH domains can rotate relative to one another and the three phosphomimetics favored more extreme rotations than WT: the T43D simulations sampled more positive twists and the S147D/T237D simulations sampled more negative twists (Fig 5 A–D). In order to highlight the range of conformations sampled for each variant, the panels in Fig 5 A–D include data from all simulations. Supplemental movies SM3, SM4, SM5, and SM6 show the range of inter-CH domain distances for the four sequences. Supplemental movies SM7, SM8, SM9, and SM10 show the range of inter-CH domain twisting for the four sequences. Landscapes for simulations initiated in either the open or closed conformation are available in the supplement (Fig S8).

TABLE 1.

C_α RMSD of MD ensembles

Genotype	Conformation	Replica	CH1	CH2	CH1 & CH2
WT	Actin-Free (4D1E)	1	0.8	0.9	1.4
	‘closed’	2	0.8	0.7	1.0
		3	0.9	0.9	1.4
		Average	0.8	0.8	1.3
	Actin-Bound (7r92)	1	1.5	1.5	3.8
	‘open’	2	1.6	2.0	3.4
		3	1.6	1.5	3.3
		Average	1.6	1.7	3.5
T43D	Actin-Free (4D1E)	1	1	0.8	3.4
	‘closed’	2	1.3	0.8	2.2
		3	1.2	0.7	1.8
		Average	1.2 *	0.8	2.5
	Actin-Bound (7r92)	1	1.6	1.6	4.0
	‘open’	2	2.4	1.5	5.5
		3	1.7	1.5	5.3
		Average	2.9	1.5	4.9 *
S147D	Actin-Free (4D1E)	1	1.0	0.8	1.3
	‘closed’	2	0.9	0.8	1.4
		3	0.8	0.8	1.2
		Average	0.9	0.8	1.3
	Actin-Bound (7r92)	1	1.5	1.7	3.6
	‘open’	2	1.6	1.6	3.5
		3	1.7	1.4	5.6
		Average	1.6	1.6	4.2
T237D	Actin-Free (4D1E)	1	1.1	0.7	1.2
	‘closed’	2	1.2	1.0	1.6
		3	1.0	0.9	1.3
		Average	1.1 *	0.9	1.4
	Actin-Bound (7r92)	1	1.7	1.8	3.3
	‘open’	2	1.4	2.1	3.4
		3	1.8	2.4	3.0
		Average	1.6	2.1	3.2

CH1 = AMBER RESIDUES :8–22,41–45,64–79,98–111

CH2 = AMBER RESIDUES :125–134,155–162,179–192,214–226

^*= two tailed equal variance students t test has p < 0.05

The CH domains for simulations initiated in the actin-bound-like conformation spontaneously twisted and came closer together to sample actin-free-like conformations. The transition from an actin-bound to actin-free like conformation involved a loss of α-helical structure in residues 144–148. The helical structure in this region connects but separates the two CH domains. Loss of helical structure allows the two domains to move closer to one another and to form inter-domain hydrophobic contacts and salt bridges (e.g., K42-D230, R53-D224). Two dimensional histograms of the inter-CH domain distance and torsion show multiple ‘hot spots’ for combinations of these two conformational metrics. The locations and sampling of these hot-spots changed as various phosphomimetics were introduced. These results indicate that, in the absence of actin, a “closed” actin-free-like conformation is favored for WT actinin. We speculate that the hydrophobic effect drives closure of the ABD and that binding of the ABD to actin stabilizes helical structure of residues 144–148. These results also indicate that the studied phosphomimetics, and the likely effect of phosphorylation at these sites, is to increase the distance between the CH domains and to accentuate the degree of twisting between the CH domains (Fig 5 A–D). There was appreciable variation among our replicate simulations initiated in the actin-bound-like conformation (Fig S9 A–D). The timescale of interdomain closure varied between 50 and 500 ns across simulation replicates (Fig S9 E–F). The extent of closure and extent of twisting also varied between the WT and phosphomimetic simulations (Fig S9 A–D). We speculate that these simulations do not completely capture a complex relationship between the interdomain distance and interdomain torsion.

The data in Fig 5 A–D span a conformational space that connects the actin-free-like and actin-bound-like states of the ABD. This series of conventional MD simulations allowed us to observe closure of the ABD without biasing the transition or the transition path (Videos SM3-SM10). Our analysis of these simulations suggests that a more “closed” or actin-free-like conformation is favored in the absence of actin, but that the phosphomimetics to better assess quantitative differences in the relative stabilities of the actin-free and actin-bound open and closed conformations, we performed umbrella sampling molecular dynamics (USMD). We analyzed the 2D distance/dihedral maps and identified a smooth path connecting actin-bound and actin-free like conformations of actinin that was sampled in all of the WT and phosphomimetic cMD simulations. For each sequence, we prepared 63 conformations along this path as the initial conformations for USMD simulations using the inter-domain distance as a reaction coordinate. Consistent with the cMD results, we found that actin-free-like states are more energetically favorable in the absence of actin, but that phosphomimetics stabilized more actin-bound-like conformations (Fig 5 E–G). Convergence of the free energy profiles calculated from USMD was assessed using block averaging in Fig S9.

DISCUSSION

Phosphorylation shifts the actin binding propensity of the α-actinin-2 actin-binding domain

In this study, AlphaFold 3 was selected as the primary modeling engine due to its demonstrably superior capabilities compared to both its predecessors (e.g., AlphaFold 2) and alternative frameworks such as RoseTTAFold. Recent advances have shown that AlphaFold 3 extends structural prediction to a broader biochemical landscape, explicitly modeling ligand-bound states, metal ion coordination, and chemically modified residues with a level of precision not previously achievable (38). Moreover, several comprehensive evaluations have underscored its enhanced performance across diverse biomolecular assemblies, including PTM proteins, multi-chain complexes, and macromolecular interfaces (56). Importantly, emerging analyses suggest that AlphaFold 3 exhibits markedly improved fidelity in capturing PTM-induced structural perturbations, outperforming conventional folding engines that typically neglect such chemical context (57). While AlphaFold 3 still faces challenges in accurately modeling certain phosphorylated or highly dynamic regions, benchmarking studies indicate that it achieves substantially higher concordance with experimental structures than previous generations, even in complex prediction settings (57, 58). Although deep-learning models remain inherently constrained by the limited PTM representation in structural databases, AlphaFold 3 currently represents the most comprehensive and context-aware framework available for modeling modified proteins and their assemblies.

In our work, we interpreted the AlphaFold 3-derived models with appropriate caution. We treated them as hypothesis-generating frameworks (rather than definitive atomic structures) and complemented them with molecular dynamics (MD) simulations and geometric validation (MolProbity) to assess stereochemical plausibility. By doing so, we mitigated the risk that small, predicted differences between WT, phosphomimetic and non-phosphorylatable variants arise solely from modelling artefacts of a single engine. Thus, our conclusions regarding phosphorylation-induced opening of the ABD and enhanced actin-binding propensity are supported by multiple independent computational layers, including MD simulations and electrostatic enthalpy ( ΔH_elec) calculations, rather than relying exclusively on a single model.

Computational tools like AlphaFold3 are useful for hypothesis generation and exploration of multiple structural models without requiring extensive computational infrastructure. AlphaFold3 has previously demonstrated good agreement with experimentally determined PDB structures (Pearson correlation 0.76–0.86) and predicted protein structures with very high confidence for over 85% of residues analyzed (plDDT > 90) or moderate-to-high confidence for ~10% of residues (plDDT 70–90) (58, 59). For example, AlphaFold3 computed the structure of the WT α-actinin-3 from Rattus norvegicus (AlphaFold DB: AF-Q8R4I6-F1), achieving a global plDDT of 84.88 and very high confidence (plDDT > 90) in the ABD, suggesting a strong likelihood of accurate predictions for WT α-actinin-2 structures. We note, however, that predictions for phosphorylated residues are less certain (~50% confidence), and therefore these models are interpreted in combination with MD simulations and experimental data rather than as definitive structures. Importantly, our MolProbity validation metrics further support the high quality of the AlphaFold 3 predictions. Across all models, phosphorylated, pseudo-phosphorylated, and alanine-substituted variants, as well as mono-, di-, and tri/tetrasubstituted forms, the MolProbity scores clustered near those of the experimentally solved 4D1E ABD, indicating low clash scores, minimal Ramachandran outliers, and well-optimized local geometry. These results highlight that even when incorporating multiple substitutions or post-translational modifications, AlphaFold3-generated models maintain stereochemical integrity comparable to experimentally resolved structures. Together with high plDDT values, these MolProbity results strengthen confidence in the structural plausibility of the predicted models and underscore AlphaFold3’s utility for exploring the structural impacts of site-specific phosphorylation and substitution events in α-actinin-2. These observations are consistent with prior reports that AlphaFold-derived structures represent highly accurate folded models but do not necessarily correspond to thermodynamic free-energy minima in explicit solvent, and small conformational relaxations during MD equilibration are expected. While ΔH_elec provides insight into electrostatic contributions at the actin interface, it does not capture other energetic or entropic terms that contribute to molecular interactions, underscoring the importance of integrating structural predictions with molecular dynamics simulations to assess conformational sampling.

In contrast, alanine substitutions preserve WT interdomain geometry and binding properties, indicating minimal structural perturbation. Multi-site phosphomimetic variants exhibit cumulative effects, with the largest interdomain distances and lowest torsion angles, suggesting cooperative modulation of ABD flexibility. These findings support a model in which phosphorylation dynamically modulates α-actinin-2 ABD conformation and electrostatic interaction surfaces in a site- and occupancy-dependent manner.

The MD simulations support conformational insights of AlphaFold 3. Our simulations show that the ABD samples a range of conformations along two principal degrees of freedom: the distance between the two CH domains and the relative rotation between the two CH domains. Our simulations indicated that the molecular factors that stabilize relative conformations include the propensity of residues 144–148 to adopt helical structure and electrostatic interactions between CH1 and CH2. These data suggest that the examined phosphomimetics favor actin-bound-like conformations of actinin and could stabilize α-actinin-2 and F-actin interactions (Fig 6). Prior single molecule in vitro work estimated that the α-actinin-4 and actin attachment lifetime is in the range of 2–20s, but this is dependent on the α-actinin isoform and the amount of mechanical load applied (61). Our study suggests that the studied phosphomimetics are likely to extend this half-life or increase the probability of actinin-actin association. Our simulations showed strong effects of the phosphomimetics on both the inter-CH domain distance and torsion. Effects of PTMs on the interdomain distance were clear to interpret: phosphomimetics promote open conformations of the ABD for all studied phosphomimetics. T43 is buried in the CH1:CH2 interface, a charge in this region is not easily buried in a hydrophobic environment. S147 is located in the CH1-CH2 loop that must uncoil during the actin-bound to actin-free transition. The negative charge at this location could form a greater number of hydrogen bonds (e.g. a salt bridge with K154) that disrupt the helical propensity of this region. T237 is in a loop that connects two helices in CH2. Asp at this position created a complex hydrogen bond network (with S179 and K239) that alters the CH1-CH2 interface structure. These behaviors were observed in both the cMD and USMD simulations. Effects of the phosphomimetics on CH1/CH2 torsion were less straightforward. All three phosphomimetics clearly altered the torsional landscape relative to the WT simulations, but there was high simulation to simulation variability in the interdomain torsions. There was also simulation to simulation variability in the timescale and conformational path away from the actin-bound-like conformation. Our USMD simulations do not fully account for this variability, because the reaction coordinate used for USMD only considered the interdomain distance and the impact of phosphorylation on interdomain torsions and the capacity of actinin to bind to actin could be explored further. Rotations of CH1 to CH2 may play a larger role in the actinin-actin complex, especially as the actinin dimer must bend to accommodate changes in lattice spacing during muscle contraction (62, 63). The data presented here show clear effects of T43D, S147D, and T237D on the structure and dynamics of the ABD. However, these models do not account for interactions with actin nor how those interactions may change during the cross-bridge cycle. Future modeling efforts should focus on the effect of PTMs on the α-actinin-actin interaction. Calponin homology domains are found in many actin binding proteins (e.g. dystrophin, utrophin, and filamin A-C) and our findings motivate future computational studies on CH domains to understand how unique sequences affect conformational sampling of the CH domain and its capacity to interact with actin. We suggest that future studies include conventional MD simulations that are at least 500 ns long (to reliably observe conformational transitions) and enhanced sampling simulations that can account for the effect of interdomain torsions. Further, our computationally generated predictions that these PTMs affect the strength and/or lifetime of the actinin-actin interaction should be evaluated experimentally.

FIGURE 6.

Working hypothesis of the redistribution of the open-closed equilibrium state of the α-actinin-2 actin binding domain by phosphorylation. (A) The location of the α-actinin-2 actin binding domain (ABD) is highlighted by a dashed box and the respective closed (left) and open (right) sates are highlighted unbound and bound to F-actin, respectively. The open state depicts the phosphorylation sites at the ABD that possibly enhance such a state. (B) The energy landscape for the equilibrium state between the closed and open states is highlighted for the WT (light blue) and the phosphomimetic α-actinin-2 ABD (magenta), showing that the phosphomimetic variants redistribute the equilibrium from overweight to marginally overweight for the closed state relative to the open state.

In muscle cells, phosphorylation is a mediator of pro-growth signals responsible for adaptive and maladaptive growth (64, 65). Mechanical stimulation from increased venous return and exercise as well as hormonal stimulation from sympathetic stimulation are examples of signaling cascades known to phosphorylate sarcomeric proteins as one of their signaling destination targets. Two well-studied examples of phosphorylation influencing sarcomere assembly include Telethonin and CapZ. Telethonin is a Z-disc protein responsible for stabilizing the titin N-terminus via its binding to two titin-N-termini. Phosphorylation of the Z-disc protein telethonin in residues S157/S161 is known to occur under basal physiologic conditions; however, adenoviral expression of a telethonin S157A/S161A mutant incapable of undergoing phosphorylation led to a disorganization of the T-tubule system as well as disrupted cytosolic Ca²⁺ reuptake (66). Another example is the actin capping protein or CapZ, which blocks actin polymerization at the (+)-barbed end. CapZ phosphorylation in response to phenylephrine stimulation exhibits increased binding-unbinding dynamics at the Z-dics suggestive of increased sarcomere growth and remodeling (67). In the context of α-actinin-2, phosphorylation could represent a pro-hypertrophic signal that further stabilizes actin and α-actinin-2 interactions to promote sarcomere assembly and muscle growth (29).

Another relevant aspect concerns the solvent accessibility of the identified phosphorylation sites. The accessibility of residues such as T43, S50, S147 and T237 within the ABD of α-actinin-2 could influence the ability of kinases and phosphatases to recognize and modify these sites. Structural studies have shown that phosphorylation acceptor residues are, on average, more solvent-exposed than non-phosphorylated counterparts, but a substantial subset resides in partially buried or conformationally hidden environments, suggesting that accessibility and conformational change play key roles in modification (68, 69). Moreover, predictions of solvent accessibility derived from deep-learning-generated structures (such as those from AlphaFold 3) correlate well with experimental exposure metrics (Pearson correlation coefficient, PCC ~0.815) and thus provide a useful proxy for estimating site accessibility (70). Inspection of our models suggests that S147 and T237 lie on loops oriented toward the solvent-exposed surface of the open ABD conformation, whereas T43 and S50 appear to reside in more constrained regions that may only become accessible during domain rearrangement. Accordingly, phosphorylation of S147 and T237 may be kinetically favorable under basal conditions, while T43 and S50 may require conformational fluctuations, potentially triggered by mechanical stress or actin-binding, to become accessible to modifying enzymes. This structural accessibility gradient could therefore contribute to a temporal or hierarchical phosphorylation pattern of α-actinin-2, linking mechanical stimuli, conformational flexibility and sarcomere assembly regulation.

Travers et al. (2013) investigated the phosphorylation of α-actinin-4 at tyrosine 265 (Y265), which corresponds to Y253 in α-actinin-2. Their structural analysis revealed that Y265 forms a hydrogen bond with serine 159 (S159). Introduction of a phosphomimetic substitution (Y265E) disrupted this interaction, as well as an additional hydrogen bond between histidine 266 (H266) and alanine 154 (A154). The Y265 residue, located within the CH2 domain, interacts with the connecting loop between the CH1 and CH2 domains, thereby stabilizing the closed conformation of the ABD. Although a 100-ns MD simulation was insufficient to capture the full opening transition of the ABD, co-precipitation assays demonstrated that the Y265E mutant exhibited strong binding to F-actin, suggesting a predominantly open ABD conformation (71). These findings support our hypothesis that phosphorylation at the CH1–CH2 interface promotes ABD opening to enhance actin binding. In contrast to the study by Travers et al. (71), our MD simulations are significantly longer and explore multiple phosphorylation and phosphomimetic sites, providing a more comprehensive view of the conformational dynamics involved.

In conclusion, this work suggests that α-actinin-2 phosphorylation at residues T43, S50, S147, or T237 facilitate the opening of the ABD conformed by the CH1-CH2 domains. This is part mediated via charge repulsion mechanisms between complementary charges found between the binding interfaces of the CH1 and CH2 domains. These results may be of relevance to understanding how cells control the assembly state of their cytoskeleton via phosphorylation signals arising from upstream mechanical and neurohumoral signaling. These results justify the need to study the role of such modifications in vitro and in vivo as well as the functional implications in other calponin homology domain family protein members such as dystrophin, utrophin, and filamin A-C among others.

Supplementary Material

SUPPORTING MATERIAL

Supplemental material contains Figures S1–S10 as Tables S1–S17, and Movies SM1–SM10.

SIGNIFICANCE.

The heart adapts to mechanical stress through compensatory muscle growth, involving the phosphorylation of sarcomeric structural proteins. Using advanced modeling techniques and simulations, we examined structural changes in α-actinin-2 upon phosphorylation at four specific sites to investigate potential effects on its conformation and actin-binding properties. We found that phosphorylation induces an opening of the protein, enhancing its ability to bind actin and thereby contributing to sarcomere stabilization and assembly. These findings suggest that phosphorylation of α-actinin-2 plays a key role in the heart’s adaptive response to mechanical stress by promoting muscle growth necessary to match cardiac function to physiological demands. Our research provides new insights into the molecular mechanisms supporting heart function and may inform future studies on cardiac disease.

DATA AVAILABILITY

Output files from MD simulations in this study are available on Zenodo at https://doi.org/10.5281/zenodo.18718680. These trajectories contain multiple TB of data and are available upon reasonable request from the UW Center for Translational Muscle Research (ctmr@uw.edu) or the corresponding author (csolis@fsu.edu).