Molecular Insights into the MLCK Activation by CaM

Abstract

Calmodulin (CaM) is a universal regulatory protein that modulates numerous cellular processes by using calcium (Ca²⁺) as the signal. In smooth muscle cells (SMC), one major target of CaM is myosin light chain kinase (MLCK), a kinase that phosphorylates the myosin regulatory light chain and thereby regulates cell contraction. In the absence of CaM, MLCK remains inhibited by its autoinhibitory domain (AID). While it is well established that CaM activates MLCK, the molecular interactions between these two proteins remain elusive due to the lack of structural data. In this work, we constructed a molecular model of mammalian CaM (mCaM) in complex with MLCK leveraging AlphaFold, published biochemical data, and protein–protein docking. The model, along with a strategic set of CaM mutants including a inhibitory variant soybean CaM isoform 4 (sCaM-4), was subject to molecular dynamics (MD) simulations. Using principal component analysis (PCA), we mapped out the transition path for the removal of the AID from the MLCK kinase domain to provide molecular basis of MLCK activation. Additionally, we established MLCK conformations that correspond to the active and inactive states of the kinase. We showed that mCaM and sCaM-4 cause MLCK to undergo the transition to the active and inactive states, respectively. Using two structural metrics, we computed the probabilities of MLCK activation by different CaM variants, which were in good agreement with the experimental data. Distributions along these metrics revealed that different inhibitory CaM variants impair MLCK activation through unique mechanisms. We finally identified molecular contacts that contribute to the MLCK activation by CaM. Overall, we report a de novo molecular model of CaM-MLCK that provides insights into the molecular mechanism of MLCK activation by CaM. The mechanism requires effective removal of the AID while preserving an active configuration of the kinase domain. This mechanism may be shared by other MLCK isoforms and potentially other structurally similar kinases with CaM-mediated regulatory domains.

Graphical Abstract

1. INTRODUCTION

MLCK is a serine/threonine kinase that plays an essential role in the regulation of cell contraction through phosphorylation of the myosin regulatory light chain (RLC).¹ The dysregulation of MLCK, either in its function or expression, is associated with diseases including pulmonary arterial hypertension.^2,3 Currently, there are four known isoforms of MLCK. MLCK1-3 are tissue-specific isoforms, namely, smooth muscle, skeletal, and cardiac MLCK.¹ MLCK4 is a newly discovered isoform expressed in cardiac muscle.⁴ In this work, we focus on smooth muscle MLCK (smMLCK).

MLCK is modulated by Ca²⁺ signaling. This ensures the tuning of cell contraction by the Ca²⁺ dynamics. Many biochemical and biophysical studies have suggested that the activation mechanism of MLCK by CaM involves the structural rearrangement of the MLCK regulatory domain induced by CaM binding (Figure 1). At rest MLCK is autoinhibited by its regulatory domain that is composed of the AID and CaM binding region (CaMBR).^1,6 The regulatory domain occupies the active site, specifically, the RLC binding site, which causes autoinhibition. As intracellular Ca²⁺ rises, CaM alters its conformation, which allows it to bind to the CaMBR. The binding of CaM to CaMBR results in the removal of the AID from the MLCK kinase domain, exposing the active site. This leads to the subsequent phosphorylation of the RLC. Phosphorylated RLC thus drives the relative movement of the thick and thin filaments through actomyosin cross-bridge formation, leading to cell contraction.

Figure 1.

Proposed activation mechanism of MLCK by CaM. (A) MLCK is autoinhibited at rest due to the autoinhibitory domain (AID) occupying its active site. (B) Elevated Ca²⁺ level causes CaM to change its conformation and bind to the CaM binding region (CaMBR) of the MLCK regulatory domain. (C) CaM-CaMBR dissociates from the MLCK kinase domain, removing the AID from the active site. (D) Myosin regulatory light chain (RLC) enters the exposed active site for phosphorylation.

Beyond the MLCK-specific regulatory elements, MLCK also contains canonical kinase motifs including the glycine-rich ATP binding loop (G1471-Q1477), the DFG motif (D1605-G1607), and the catalytic HRD motif (H1583-D1585) (Figure 2A).⁷ D1605 at the DFG motif coordinates with magnesium (Mg²⁺), which in turn coordinates with the pyrophosphate backbone of ATP.⁸ Together, the ATP binding loop and DFG motif lock ATP in place for hydrolysis. The configuration of the DFG loop is, therefore, indicative of the active/inactive state of several kinases. Namely, the ”DFG-in” configuration is characteristic of an active kinase, whereas the ”DFG-out” configuration indicates an inactive kinase. The HRD motif (HLD in MLCK) contains a catalytic D1585 that functions as a base to deprotonate the serine/threonine residue of the substrate. The deprotonated residue subsequently performs a nucleophilic attack on the adenosine triphosphate (ATP) to complete phosphorylation. In addition to the structural elements for ATP binding and catalysis, MLCK also has two glutamate residues (E1545 and E1589) that bind both the RLC and AID.^9,10

Figure 2.

Sequence and structure comparison of kinase homologues. (A) Sequence alignment of smMLCK, skMLCK, CaMKII, and twitchin kinase. Numbering is based on the human MYLK gene sequence (UniProt Q15746). The sequence identities of the kinase domain and regulatory domain are listed in panel B. The sequence of smMLCK is color coded based on different domains and active site residues. Regions where the CaM N-terminus domain interacts with smMLCK identified as gained or lost contacts in this work are highlighted in green and red boxes, respectively. In these regions, skMLCK residues that are not conserved with smMLCK are highlighted in red. (B) Comparison of the structures of smMLCK predicted by AlphaFold (accession code: Q15746), skMLCK predicted by AlphaFold (accession code: Q9H1R3), CaMKII (PDB entry 2VN9), and twitchin kinase (PDB: 3UTO). These kinase homologues share highly similar domain structures. The regulatory domain (AID + CaMBR) is colored in red. (C) Active site arrangement of smMLCK. The active site motifs are color coded and labeled as described in panel A. (D) Comparison of the smMLCK active site availability with and without the regulatory domain. The cavity of the active site is predicted by CavitOmiX (developed by Innophore) and colored in green.

Despite sharing an AID-based activation mechanism similar to that of other AID-reliant proteins such as calcineurin (CN), MLCK possesses distinct regulatory domain arrangements. We hypothesize that this difference leads to unique molecular interactions with CaM that enable activation of MLCK through the removal of the AID. However, these molecular interactions remain elusive, as there are limited data available for the structure of MLCK and the CaM-MLCK interactions, which limits the ability to test our hypothesis. We therefore sought to probe these molecular interactions between MLCK and the mammalian CaM (mCaM) by integrating published experimental data and molecular simulations data. Towards this end, we first built a molecular model of mCaM in complex with the smMLCK kinase domain using biochemical data and protein–protein docking. The molecular model, along with a strategic set of CaM mutants, was simulated, and a transition path of the AID from the MLCK kinase domain was identified. To quantify the MLCK activation likelihood, we defined structural metrics that are correlated with MLCK activation: the distance between CaM-CaMBR and the MLCK kinase domain and the distance between the ATP binding DFG motif and the catalytic HRD motif. Distributions along these metrics revealed that impaired MLCK activation by the CaM variants occurs through different mechanisms. The MLCK activation probabilities computed are in good agreement with published data. Finally, we identified interactions between CaM and MLCK that contribute to the MLCK activation to propose molecular sites for future mechanistic studies via mutagenesis. Altogether, our study indicates that the molecular mechanism of MLCK activation by CaM entails effective removal of the AID while preserving an active configuration of the kinase domain. This mechanism may be shared by other kinase homologues that utilize CaM binding regulatory domains. This work provides insights into the molecular basis of MLCK modulation by CaM, which could be leveraged for the rational engineering of CaM variants to mitigate MLCK-associated diseases such as pulmonary arterial hypertension,² atherosclerosis, pancreatitis, and asthma.¹¹

2. MATERIALS AND METHODS

2.1. Structures and Systems.

Structures used and systems simulated in this study are summarized in Tables S1 and S2.

2.2. In Silico Mutagenesis and Relaxation.

To create CaM mutants in silico, we first stripped the side chain atoms of the residue to mutate. Residue name was then changed to the mutated residue in the PDB file. Missing side chains were repaired using TLEAP. We further relaxed the mutated structure of CaM E14A/T34K/S38 M using the following procedure. To parametrize the protein atoms, the ff14SB¹² force field was used. The mutated structure was then solvated using TLEAP in a TIP3P¹³ waterbox (distance from the protein to the box edge: 12 Å) and 150 mM KCl to simulate a physiological cytosolic environment. Relaxation was next performed following the protocol described previously.¹⁴ All simulations were performed using Amber 18.¹⁵ A nonbonded cutoff of 10 Å was applied. SHAKE¹⁶ was used to constrain bonds involving hydrogen, allowing for a 2 fs time step to be used for simulations. The system was first minimized using the steepest descent for the first 200 cycles and switched to conjugate gradient with a maximum number of 50,000 cycles. During minimization, a positional restraint of 10 kcal mol⁻¹ Å⁻² was imposed on all of the protein backbone atoms. The following steps were carried out in a Langevin thermostat with a collision frequency of 3 ps⁻¹: in the first step, the system was heated from 0 to 300 K in an NVT ensemble with a positional restraint of 10 kcal mol⁻¹ Å⁻² imposed on all protein backbone atoms. In the next step, the system was heated from 0 to 300 K in an NPT ensemble with the positional restraint imposed on all protein backbone atoms (3 kcal mol⁻¹ Å⁻²) except the mutated residue backbone. Following the heating stages, the system was equilibrated in an NPT ensemble for 1 ns, with the positional restraint further reduced to 1 kcal mol⁻¹ Å⁻². Finally, productions runs were performed in an NPT ensemble with randomized initial velocities in triplicate.

2.3. Rosetta Comparative Modeling of sCaM-1 and sCaM-4.

The structures of sCaM-1 and sCaM-4 bound to the CaMBR were constructed using Rosetta¹⁷ following a similar protocol described previously.¹⁴ Flags and input files (”rosetta_cm_flags” and ”rosetta_cm.xml”) can be found in the repository posted below. The highest scored structure was used for relaxation. Built structures were subject to relaxation following steps described in section 2.2. These structures were simulated for 2 μs with randomized initial velocities in triplicate. Clustering analysis was performed using CPPTRAJ, and the most populated structure was used for subsequent simulations.

2.4. CaM-MLCK Complex Modeling.

To build a complete mCaM-MLCK model comprising mCaM, the MLCK kinase domain (V1460-M1720), the AID (K1721-A1741), and the CaMBR (R1742-S1760), we combined protein–protein docking and loop modeling. Specifically, we first used protein–protein docking to predict the binding mode between CaM-CaMBR and the kinase domain predicted by AlphaFold. We then performed loop modeling to fill in the missing AID between the MLCK kinase domain and CaMBR in the predicted poses. ClusPro¹⁸ was used to perform protein–protein docking with CaM-CaMBR as the ligand and the kinase domain as the receptor. The mCaM triple mutant E14A/T34K/S38 M was first mutated and relaxed, as described above. This triple mutant was used in the docking. We additionally specified mCaM E14, T34, and S38 to be ”attraction residues” (residues that are assumed to participate in the binding) in the ClusPro setup. Generated poses were screened based on two criteria: (1) if mCaM is positioned close to the MLCK active site, i.e., the binding of mCaM is preventing substrate entry; (2) if the distance between the MLCK C-terminus and CaMBR N-terminus is <60 Å, i.e., the spatial distance is not too long for the insertion of the AID. The mCaM triple mutant was replaced with mCaM before loop modeling. RosettaRemodel¹⁹ was then used to model the AID. Flags and input files (”rosetta_remodel_flags” and ”missing.remodel”) for RosettaRemodel can be found at the repository posted below. The highest-scoring structure was used for subsequent simulations.

To simulate CaM variants in complex with MLCK, the variant structures of CaM were superimposed on the mCaM structure. Coordinates of mCaM were removed, creating complexes of MLCK and different CaM variants. These complexes were used for subsequent simulations.

2.5. MD Simulations and Analyses.

All-atom simulations of the CaM-MLCK complex and CaM-free MLCK were performed following steps described in section 2.2. Long simulations of 2 μs in triplicate and short simulations of 250 ns in 50 replicates were excecuted. Simulations were initiated from the equilibrated configuration with randomized initial velocities. In these simulations, Mg²⁺ and ATP were not included as it is demonstrated that mCaM-MLCK interactions are independent of ATP binding and catalysis.²⁰

RMSD, clustering, distance, $R_{\mathrm{g}}$, contact, and PCA analyses were performed using CPPTRAJ. To investigate the effect of CaM variants on the conformational distributions of the CaM-MLCK complex and MLCK kinase domain, we performed PCA analysis. Specifically, a Cartesian coordinate-based covariance matrix was constructed using the mCaM trajectories from the 250 ns simulations. Next, trajectories of CaM variants were aligned to the mCaM structure and projected along the eigenvectors of the matrix. The probability densities of the principal component distributions were converted to free energies using the Boltzmann inversion, $G\hspace{0.17em}=\hspace{0.17em}-k_{\mathrm{B}}T\hspace{0.17em}\ln p$, where G is the free energy, $k_B$ is the Boltzmann constant, $T$ is the temperature, and $p$ is the probability density.

3. RESULTS AND DISCUSSION

3.1. Predicted smMLCK Is Structurally Similar to Its Kinase Homologues.

Currently, there are no structural data available for smooth muscle MLCK (smMLCK). To study the structural dynamics of smMLCK, we first used AlphaFold^21,22 to obtain predicted structures of the kinase domain and regulatory domain (AID + CaMBR) of smMLCK (accession code: Q15746). For comparison, we also retrieved the AlphaFold predicted structure of skMLCK (accession code: Q9H1R3), which has 50% sequence identity to smMLCK in its kinase and regulatory domains. The superposition of the predicted MLCK structure with other crystallographically resolved kinases, such as twitchin kinase (PDB: 3UTO) and Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) (PDB: 2VN9), showed that these homologues are structurally similar (Figure 2B). The kinase homologues share the canonical kinase domain configuration, including the glycine-rich ATP binding loop, DFG motif, and catalytic HRD motif (Figure 2C).⁷ Specific to smMLCK, E1546 and E1590 are involved in the binding of the RLC and AID. The major differences among the kinase homologues are in the regulatory domain consisting of the AID (K1721-A1741) and CaMBR (R1742-S1760). Notably, in smMLCK, the AID is bound between two helices. In these regions, E1553 is in contact with K1737, consistent with a previous finding that showed that rabbit smMLCK DED784-786 (equivalent to DED1552-1554 in human smMLCK) are AID-binding residues.⁹ Similarly, in CaMKII, a recent report established the important role of S280 in anchoring the AID to modulate CaMKII function.²³ Altogether, these AID-kinase domain interactions are critical for the autoinhibition of the kinase, such that their disruption could activate the kinase. Under physiological settings, such disruption occurs through the binding of CaM. We additionally observe that the AID is in the proximity of the two RLC binding glutamates, E1545 and E1589. We next used CavitOmiX to compute the pockets in the structure as a measure of the active site availability.²⁴ The availability is shown as green isosurfaces, which depict the volume of the active site pocket in the structure. By comparing the active site availability (green isosurface) of smMLCK in the presence and absence of the regulatory domain (Figure 2D), we noted that AID partially occupies the RLC binding site that is otherwise solvent accessible when AID is removed. A similar regulatory domain arrangement is observed in skMLCK. The regulatory domain of CaMKII is positioned more deeply in the active site. In contrast, the regulatory domain of twitchin kinase is deeply buried in the active site, completely blocking substrate entry. Overall, the structural similarity with kinase homologues and the regulatory domain configurationin are in line with experimental data and lend us confidence in using the predicted smMLCK structure for subsequent molecular modeling and simulations.

3.2. smMLCK Regulatory Domain Favors Binding to the Active Site.

Several biochemical and biophysical studies reveal that MLCK is autoinhibited by its regulatory domain (AID + CaMBR) occupying the active site, preventing the RLC from binding.⁵ To validate the AlphaFold predicted structure, we first simulated interactions between the regulatory domain and kinase domain through MD simulations. As shown in Figure 3A, the average number of contacts between the kinase domain and regulatory domain in the last 200 ns of the simulations increases relative to the first frame, indicating more interactions. These interactions are stable over simulations of 1 μs (Figure S1A). We also quantified the solvent accessible surface area (SASA) of the kinase domain as a measure of active site availability. We observe that SASA decreases (Figure 3A and Figure S1B), indicating that the active site became less solvent accessible. This correlates with the increase in the number of contacts, indicating the binding of the regulatory domain to the active site. Mapping of the contacts identified in the contact map to the structure revealed that the binding interface is localized to the RLC binding groove and active site (Figure 3B, cyan). The C-terminus of the regulatory domain collapsed toward the active site and formed stable interactions. It is noteworthy that R1743 in the regulatory domain interacts with E1545 and E1589 at the active site, consistent with previous findings that basic residues at the regulatory domain and acidic residues at the active site are involved in the binding of the AID.^10,25 In addition, D1666 in the kinase domain is in close proximity to R1753 and R1757 of the CaMBR and therefore forms salt bridges. Despite the collapse of the regulatory domain, the ATP binding site remained accessible, as evidenced by a cavity detected with CavitOmiX (gray isosurface). Within this cavity are structural elements crucial for ATP binding: the DFG motif and ATP binding loop. This is consistent with previous findings showing that the binding of ATP is independent of the presence of the regulatory domain.^26,27 Altogether, these results validate the smMLCK model predicted by AlphaFold and suggest that, in the absence of CaM, the regulatory domain favors binding to the active site of MLCK.

Figure 3.

MD simulations of the MLCK kinase domain with the AID and CaMBR in the absence (A, B) and presence of CaM (C–E). (A) Relative changes of the number of contacts between the kinase domain and regulatory domain as well as SASA of the kinase domain from the last 200 ns of the simulations relative to the first 200 ns of the simulations. (B) Contact map analyses revealed interactions between the kinase domain and regulatory domain. Representative structures indicate that the regulatory domain binds to the MLCK active site. Contacts at the kinase domain are highlighted in cyan. Dashed squares show the salt bridges formed. CavitOmiX indicates that the ATP binding site remains available (gray isosurface) after AID binds to the active site. (C) Predicted smMLCK (blue) and CaM (green) complex structure. The MLCK regulatory domain is colored in red. In addition, the AID position in the absence of CaM is rendered transparent for comparison. (D) Left: the CaM-MLCK complex before MD simulation. Right: in one of the MD simulations, CaM dissociates from MLCK, leaving the active site (white) open and allowing for substrate entry. (E) RMSD of the three replicates of the CaM-smMLCK and CaM-CaMBR complexes. For the CaM-MLCK complex, the RMSD of one of the replicates increased after 100 ns, indicating the removal of the AID from the MLCK kinase domain. In contrast, the RMSD of the CaM-CaMBR complex remained stable.

3.3. AID Removal Correlates with MLCK Active Site Rearrangement.

3.3.1. Modeling the CaM-MLCK Complex.

It was hypothesized that CaM activates MLCK by binding to the regulatory domain (AID + CaMBR) and removing it from the active site, thus allowing entry of the RLC (Figure 1).²⁸ We postulated that, in order for CaM to remove the AID, CaM first would have to bind to the regulatory domain and specifically the CaMBR. This binding forms a transient complex comprising the CaM-CaMBR-MLCK kinase domain, where CaM-CaMBR temporarily blocks the MLCK active site. As the CaM-CaMBR moves away from the active site, it pulls the AID along, exposing the active site for substrate binding and catalysis (Figure 1). Since there are no structural data available for the mCaM-CaMBR-MLCK kinase domain complex, we first predicted the binding mode between CaM-CaMBR and the MLCK kinase domain (predicted by AlphaFold) using ClusPro,¹⁸ a protein–protein docking algorithm. We applied several constraints based on published biochemical data on MLCK activation by mCaM. VanBerkum et al. established that a mCaM triple mutant E14A/T34K/S38 M disrupted MLCK activation yet did not perturb the binding of mCaM to CaMBR.²⁹ We postulate that the triple mutant disrupted MLCK activation due to its stronger binding relative to the WT, preventing the removal of AID. Therefore, we first created in silico the mCaM triple mutant bound to CaMBR and performed a 1 μs MD simulation to relax the mutated structure. Consistent with experimental observations, the CaM triple mutant remained bound to CaMBR in the canonical ”wrapped around” configuration. With the refined mCaM structure, we used ClusPro to predict the binding pose between the MLCK kinase domain and CaM-CaMBR triple mutant. The ClusPro predicted poses were next screened based on two criteria: (1) whether mCaM is blocking the MLCK active site, either partially or completely, and (2) whether the distance between CaMBR and the C-terminus of MLCK is suitable for AID insertion. The first criterion is based on the assumption that, after binding to the regulatory domain, CaM-CaMBR temporarily blocks the active site. It is the subsequent separation of CaM-CaMBR from the kinase domain that leads to the exposure of the MLCK active site. The second criterion was applied to ensure that there is enough space for AID insertion. We then replaced the mCaM triple mutant with mCaM in the predicted and screened poses. We lastly used Rosetta to model the missing AID and generated a complete model consisting of mCaM, the MLCK kinase domain, AID, and CaMBR (Figure 3C). In this model, CaM-CaMBR interfaces with the MLCK active site and prevents RLC entry. We additionally observed that the AID is displaced from the active site, in contrast to the AID position observed in the AlphaFold predicted structure (Figure 3C, transparent helix).

3.3.2. MD Simulations.

To determine the molecular interactions between mCaM and MLCK, the predicted structure of the CaM-MLCK complex was subjected to 2 μs all-atom MD in triplicate simulations. Interestingly, we noted that, in one of the replicates, CaM-CaMBR moved away from the MLCK active site (white surface), causing the removal of the AID and exposing the active site for potential RLC binding (Figure 3D). RMSD analyses of the CaM-MLCK complex indicated that AID removal occurred early on in the simulation (around 100 ns, Figure 3E top panel). In contrast, the RMSDs of CaM-CaMBR in all three replicates were similar and stable, indicating no significant conformational change occurred. While we observed AID removal, it happened in only one out of three replicates. Therefore, to sample the molecular details of the AID removal more robustly and collect additional statistics, we performed short simulations of 250 ns using 50 replicates. We next used PCA to evaluate the conformational distributions of the CaM-MLCK complex and only the MLCK kinase domain (Figure 4). As a dimensionality reduction analysis tool, PCA was expected to reveal global structural changes in the protein conformation. Previously, we have successfully applied this method to dissect the ion-binding states of S100A1.³⁰ We first constructed a coordinate covariance matrix of the mCaM-MLCK complex using trajectories from 50 replicate short simulations. With the covariance matrix, the principal components identified account for the largest variances in the protein coordinates, i.e., the largest structural changes. Here, we focus on PC1 and PC2, the first two principal components that account for 48 and 22% of the total coordinate variance, respectively. The resulting matrix therefore describes the conformational change of the CaM-MLCK complex during the simulations in terms of the principal components. To determine the extent to which the CaM variants adopt conformations similar to those of mCaM, trajectories of different variants were projected onto the PCA bases.

Figure 4.

Conformational distributions revealed by PCA analysis. Trajectories from 50 replicate short simulations were used to construct PC matrices, with which trajectories of different CaM variants were projected. To obtain the free energy landscapes along PC1 and PC2, we computed the joint probability distribution along both PCs and converted the probability to free energy using the Boltzmann inversion at 310 K. (A) Conformational distributions of the CaM-MLCK complex. Left: the free energy landscape from the 2 μs simulations of the mCaM-MLCK complex. Five low energy areas were identified (green asterisks). Area 1 was identified as the predominant low energy area, whereas areas 2 to 5 are intermediate energy areas. Right: the transition path of the MLCK regulatory domain from the kinase domain. The structure of CaM is hidden for better visualization of the path. (B) Free energy landscapes from the 250 ns simulations of MLCK in complex with mCaM, K30E, sCaM-1, and sCaM-4. (C) Conformational distributions of the MLCK kinase domain. Left: the free energy landscape from the 2 μs simulations of the mCaM-MLCK complex and CaM-free MLCK combined. Three low energy areas were identified (green asterisks). Representative structures of these areas are shown with the N lobe of the kinase domain in blue and the C lobe of the kinase domain in white. (D) Free energy landscapes from the 250 ns simulations of MLCK in complex with mCaM, K30E, sCaM-1, and sCaM-4. In the 250 ns simulations, MLCK samples three low energy areas (blue asterisks) slightly different from those identified in the 2 μs simulations (green asterisks).

We next computed the joint probability distribution along PC1 and PC2 and converted the probability density to free energy using Boltzmann inversion at 310 K ($G\hspace{0.17em}=\hspace{0.17em}-k_{\mathrm{B}}T\hspace{0.17em}\ln p$ where $\mathrm{G}$ is the free energy, $K_B$ is the Boltzmann constant, $T$ is the temperature, and $p$ is the probability for a given point). These free energies were used to construct the free energy landscapes. For each system, the free energy landscape depicts the conformational distribution of the CaM-MLCK complex during the simulations. In low energy areas, conformations are sampled more frequently. As shown in Figure 4A (left panel), for the mCaM-MLCK (2 μs x 3) simulations, one major low energy area (area1) was identified, as well as several intermediate energy areas (areas 2 to 5). To visualize the structural changes PC1 and PC2 describe, we mapped the CaM-MLCK complex structure onto both PCs (Figure S2A). The PC arrows indicate both the direction (arrowhead) and the magnitude (arrow length) of the structural change. Taken together, PC1 and PC2 describe the removal of AID from the MLCK kinase domain. To examine the path of the AID removal, we obtained representative structures from areas 1 to 5 by performing clustering analysis of the simulation snapshot frames that fall within a given low energy area. We confirmed that these structures depict the removal of the AID from the active site (Figure 4A, right panel). Taken together, the displacement of the regulatory domain follows the path outlined in the free energy landscape, from area 1 to area 5 through areas 2, 3, and then 4. Due to the short simulation time (250 ns versus 2 μs), the short simulations (Figure 4B, ”mCaM”) sampled only a part of the transition path from area 1 to area 3 through area 2. It should be noted that, while no contour is shown at the area 3 asterisk, there are conformations distributed around that point.

We next examined the conformational distribution of the MLCK kinase domain. A similar workflow was applied with the exception that only the kinase domain coordinates were used for PCA matrix construction and projection. This allows us to separate the local kinase domain conformational changes from the overall CaM-MLCK complex conformational changes. Projection of the MLCK kinase domain onto the first two principal components (29 and 13% of total variance) revealed relative motions between the two lobes: the N-terminus ATP binding lobe and C-terminus RLC binding lobe (Figure S2B). To construct a free energy landscape that encompasses conformations sampled in both the absence and presence of mCaM, we combined the trajectories of the long simulations of both the CaM-free MLCK and mCaM-MLCK complex. These trajectories were projected onto the covariance matrix of the kinase domain constructed using trajectories of short simulations of mCaM-MLCK. The free energy landscape indicates that the kinase domain conformation is predominantly localized to area 3 (Figure 4C). Without mCaM (CaM-free MLCK simulations), the kinase domain conformation transitions to area 2. In contrast, with mCaM (mCaM-MLCK simulations), the kinase domain conformation shifts toward area 1. The superposition of the representative structures from these areas indicate that the area 1 conformation is more compact between its N (ATP binding) and C (RLC binding) lobes than conformations in the other areas (Figure S3A). This is further validated by the radius of gyration ($R_{\mathrm{g}}$). As shown in Figure S3B, in area 1, MLCK exhibited a lower $R_{\mathrm{g}}$, indicating a more compact structure, whereas in area 2, MLCK has the highest $R_{\mathrm{g}}$, indicating a more extended structure. The compactness of area 1 conformations is consistent with previous studies that discovered that MLCK became more compact upon CaM activation.^31,32 Additionally, in area 1 MLCK exhibits a configuration with a short distance between D1605 of the Mg²⁺-coordinating DFG motif and D1585 of the catalytic HRD motif (Figure S3C). Taken together, we speculate that the area 1 conformation resembles an active state of MLCK while the area 2 conformation resembles an inactive state, as judged by its more extended conformation and longer distance between D1605 and D1585. Conformational landscapes of the short simulations of mCaM-MLCK revealed that MLCK samples two major states (areas 1 and 2 in blue). Moreover, these states fall on a path transitioning from the predominant state (area 1 in blue) toward the active state (area 1 in green), as shown in Figure 4C.

3.4. Plant CaM-Derived Variant Simulations Revealed CaM-MLCK Interactions for MLCK Activation.

3.4.1. RMSD.

To gain molecular insights into the MLCK activation mechanism, we performed simulations of MLCK bound to a series of experimentally characterized CaM variants, including soybean CaM-1 (sCaM-1), sCaM-4, K30E, G40D, M36I, T34K, and ΔNCaM. Van Lierop et al. showed that sCaM-4 exhibited significantly reduced MLCK activation as compared to mCaM, while sCaM-1 showed comparable activity.⁵ Sequence alignment suggested that K30, M36, and G40 might be the source of the difference in MLCK activation. Further mutagenesis (K30E, M36I, G40D) studies indicated that K30 and G40 contributed to the reduced MLCK activation by sCaM-4 while M36I did not impact MLCK activation. In addition to the plant isoform-derived variants, we considered ΔNCaM, a CaM variant lacking N-terminal residue 2–8 (DQLTEEQ), which is only able to partially activate smMLCK (50%).³³ Structural studies using small-angle X-ray scattering (SAXS) indicated that this reduced activation was due to the inability of CaM to remove the AID from MLCK.²⁸ Lastly, we also included T34K, one of the mutations in the triple mutant used in the protein–protein docking workflow.

We first assessed the RMSD as a measure of the overall structural fluctuation (Figure S4). Presented here is the distribution of the RMSD of the CaM-MLCK complex from 50 replicates for each system. Results from the long simulations (Figure 3E, upper panel) indicate that the AID removal corresponds to CaM-MLCK RMSD values above 3 Å. As an approximate metric to quantify AID removal from MLCK, we applied a cutoff of 3 Å and computed the probability densities with RMSD above the cutoff from the 50 replicates. Bootstrapped averages and standard deviations of the 50 replicates (500 draws) were then calculated by using these probability densities. As shown in Figure S4B, most of the variants have similar probability densities as mCaM.

3.4.2. PCA.

To assess the conformational distributions of the CaM-MLCK complex and MLCK kinase domain in the presence of different CaM variants, we projected the CaM variant trajectories onto the PCA bases constructed using the mCaM trajectories as described in section 3.3.2.

The PCA of the CaM-MLCK complex indicates that all variants showed different degrees of AID removal (Figure S5A) along the path in Figure 4A. Conversely, K30E and sCaM-4 present greater extents of AID removal, as characterized by the wide span of their energy landscapes along both principal components. Interestingly, these two variants seem to have shifted the transition path. Specifically, instead of going from area 1 to area 5 through areas 2, 3, and then 4 as illustrated in Figure 4A, these two variants undergo transitions from area 2 to areas 3, 4, and 5. Interestingly, sCaM-1 also exhibited an altered transition path, albeit in an opposite direction.

We next examined the free energy landscapes of the MLCK kinase domain conformations. The conformations of the kinase domain when bound to mCaM predominantly distribute between areas 1 and 2. In contrast, conformations with the variants, especially K30E, T34K, and sCaM-4, migrated from area 1 toward area 3 (Figure S5B). As a reference, we labeled the low energy areas from the 2 μs simulations of the mCaM-MLCK complex (area 1) and CaM-free MLCK (area 2) in green, which resemble the active and inactive states of MLCK, respectively. Taken together, starting from area 1, MLCK in the presence of mCaM transitioned toward area 2, which resembles an active state. Conversely, MLCK, when complexed with other CaM variants, migrated toward area 3, which is structurally similar to an inactive state. Meanwhile, sCaM-1 resulted in conformations not sampled in the mCaM simulations.

To visualize the structural differences induced by mCaM versus sCaM-4, we superimposed the structures of the MLCK kinase domain in complex with mCaM from area 2 (of MLCK PCA) and that in complex with sCaM-4 from area 3 (of MLCK PCA) (Figure 5A,B). The kinase domain overall is more compact with mCaM bound and more extended with sCaM-4 bound (Figure S6). One consequence of the more extended structure is that the ATP binding loop is shifted farther away from the catalytic residue D1585, which is unfavorable for phosphoryl transfer (Figure 5B). More importantly, the DFG motif in the sCaM-4 case flipped from the ”DFG-in” configuration (as seen in mCaM) to one that is pointing outward, suggesting inactivation of MLCK. While it does not fully present the ”DFG-out” state, it is clear that D1605 is shifting outward from the active site.

Figure 5.

CaM-MLCK distance vs DFG-HRD distance. (A) Superposition of MLCK in complex with mCaM from MLCK PCA area 2 and that with sCaM-4 from MLCK PCA area 3. mCaM is colored in blue, sCaM-4 in red, and MLCK in white. The arrow indicates the removal of the AID from the MLCK kinase domain. (B) MLCK active site when complexed with mCaM (blue) or sCaM-4 (red). The arrow indicates the distance between D1585 and D1605. The DFG motif shifts from a ”DFG-in” configuration (mCaM) to a ”DFG-out” configuration (sCaM-4). (C) The two distance pairs were used to define the active state of MLCK. The joint probability densities along both distance pairs were converted to free energies using the Boltzmann inversion. (D) MLCK activation was approximated using probability densities of the distance distributions satisfying CaM-MLCK and DFG-HRD distance metrics. The averages and standard deviations were computed from 50 simulation replicates using bootstrapping with 500 draws. *p <0.05.

3.4.3. Structural Metrics that Define MLCK Activation.

To quantify the extent of MLCK activation by CaM, we propose two structural metrics that are based on the hypothesis that MLCK activation consists of AID removal without compromising the active site integrity. These two metrics are (1) the CaM-MLCK distance (distance between the centers of mass of CaM/CaMBR and the MLCK kinase domain) as a measure of the AID removal and (2) the distance between D1605 of the DFG motif and D1585 of the HRD motif (distance between the two carboxylate carbon atoms of D1605 and D1585). We then defined thresholds for these metrics to classify the active state of MLCK: (1) CaM-MLCK distance >30 Å, and 2) D1605 and D1585 distance <5 Å. The probability densities in areas that satisfy these arbitrary thresholds relative to the total area were computed as a proxy of the probability of MLCK being activated by the CaM variants. As shown in Figure 5D, mCaM and sCaM-1 both present high probabilities of MLCK activation. In contrast, other variants exhibit substantially lower probabilities. Overall, this is consistent with the experimental observations that sCaM-1 showed comparable MLCK activation relative to mCaM whereas sCaM-4, K30E, G40D, T34K, and ΔNCaM exhibited reduced MLCK activation. The only exception is M36I. In our model, M36 is located close to the C-terminus, which usually undergoes high structural fluctuations during simulations. Therefore, this limitation of our model might be an artifact and exaggerates the effect of M36I on the MLCK configuration, resulting in the inconsistency. Interestingly, the distributions along the two structural metrics revealed that K30E, G40D, T34K, ΔNCaM, and sCaM-4 all led to reduced MLCK activation probabilities, albeit via different mechanisms (Figure S7C). For example, while K30E and sCaM-4 demonstrated greater AID removal, they also distorted the MLCK kinase domain. In contrast, the other variants did not fully remove the AID but distorted the kinase domain relative to mCaM.

3.4.4. CaM-MLCK Contacts.

To identify the molecular interactions that contribute to MLCK activation, we computed contact maps of interacting residues. The PCA of the MLCK kinase domain revealed that, when mCaM is bound, the CaM-MLCK complex’s conformations shift from area 1 to area 2. In contrast, the conformations shift from area 1 to area 3 when bound to K30E and sCaM-4. We compared the mCaM-MLCK contacts between area 1 and area 2, K30E-MLCK contacts between area 1 and area 3, and sCaM-4-MLCK contacts between area 1 and area 3 (Figure 6). Contacts that are lost upon transition from area 1 are colored in red, whereas those that are gained are colored in green. These difference contact maps provide a straightforward visualization of the change in contacts that correlate with the conformational transition. Notably, when the MLCK kinase domain moves from area 1 to area 2 when bound to mCaM, specific contacts were gained (Figure 2A green box, and Figure 6A green dashed boxes). In contrast, these contacts are lost when the kinase domain moves to area 3 when bound to K30E and sCaM-4. We speculate that these contacts likely contribute to the divergent activation pathways of mCaM, K30E, and sCaM-4. These contacts are localized to the N-terminus domain of CaM, and they are predominantly hydrophobic. This might explain why the ΔNCaM mutant showed significantly reduced activation of MLCK, as hydrophobic interactions were lacking. Interestingly, we noted that, in these regions, skMLCK is slightly more hydrophobic than smMLCK, which might correlate with the more dramatic effect of ΔNCaM on skMLCK activation than smMLCK.

Figure 6.

Contacts that contribute to different MLCK states. (A) Difference contact maps between mCaM and MLCK from MLCK PCA area 1 vs area 2, K30E and MLCK from MLCK PCA area 1 vs area 3, and sCaM-4 and MLCK from MLCK PCA area 1 vs area 3. Contacts gained relative to area 1 are colored in green, and contacts lost relative area 1 are colored in red. Green dashed boxes highlight contacts in mCaM that are gained in contrast to those in K30E and sCaM-4 that are lost. Red dashed box highlights contacts in mCaM that are lost to a greater extent relative to K30E and sCaM-4. (B) Contacts in green dashed boxes in panel A are visualized and represented as van der Waals radii spheres. MLCK kinase domain is colored in blue, mCaM in green, and the regulatory domain in red. (C) Local contacts between mCaM and the MLCK kinase domain. The dotted lines denote hydrogen bonding between K13 and the backbone carbonyl oxygen atoms of P1660 and M1662.

Conversely, specific contacts were lost as the MLCK kinase domain transitions from area 1 to area 2 when bound to mCaM to a greater extent than K30E and sCaM-4 (Figure 2A red box and Figure 6A red dashed box). These contacts are mostly electrostatic and are localized to the N-terminus domain of CaM. Taken together, the gained contacts allow CaM-CaMBR to ”roll off” the MLCK kinase domain, leading to the loss of the other contacts and eventually removal of the AID. It is noteworthy that K30 is located at the lost contact site and in the proximity to E1548. Upon mutation to Glu, K30E causes it to electrostatically repel from E1548 and this explains why the binding of CaM K30E and sCaM-4 resulted in distortion of the MLCK kinase domain (Figure S9A,B). In addition, T34 is located close to E1553. Mutation of T34 to Lys results in electrostatically attractive interactions that ”glue” CaM to MLCK. Forced removal of the AID by CaM then distorts the MLCK kinase domain (Figure S9C,D).

3.5. Limitations.

In this work, we predicted the binding mode between CaM and MLCK based on mutagenesis data and protein–protein docking. Through simulations of MLCK bound to different CaM variants, we gained molecular insights into the mechanism of MLCK activation by CaM. However, we also recognized that the molecular model we built would need further validation. In addition, a part of the molecular model (the MLCK kinase domain) is an AlphaFold predicted structure and therefore might not represent the actual structure of MLCK. Hence, we propose a set of mutation sites (mCaM: K13A/E, S17L, R37A/E, smMLCK: L1658E, S1659D/K, and M1662F) that could help validate and refine the model through experimental approaches such as MLCK kinase activity assays and CaM-MLCK binding assays.

4. CONCLUSIONS

This work provides molecular insights into the mechanism of MLCK activation by CaM, which offers directions for future exploration of the molecular mechanism and aids in understanding the regulation of MLCK function and cell mechanics. We tested the hypothesis that the mechanism of CaM-dependent MLCK activation entails CaM removing the AID from MLCK by binding to the regulatory domain and dissociating from the kinase domain. Toward this end, we constructed a molecular model of CaM in complex with MLCK based on published biochemical data. The model was simulated in the presence of different CaM variants from which we mapped out a path of AID removal from the MLCK kinase domain to activate the kinase. We found that MLCK adopts different conformations in the presence of different CaM variants. We postulate that these conformations correspond to the active and inactive states of the kinase. Quantification of MLCK activation showed that our model and simulation results are in agreement with the published data. A plethora of interactions that contribute to the MLCK activation was identified, which could be targets for mutagenesis to challenge our hypothesis (see below). We propose that the molecular mechanism of MLCK activation by CaM entails the effective removal of the AID while preserving an active configuration of the MLCK kinase domain. Therefore, there could be multiple strategies to engineer CaM to regulate MLCK function, i.e., by interfering with the AID removal, or by distorting the MLCK kinase domain. This is analogous to our recent finding that CaM mutations could lead to inhibition of CN via different mechanisms (e.g., Ca²⁺ binding, AID removal).³⁴

Based on the sequence similarities at the identified contact sites, this mechanism of kinase activation may be shared by other CaM-regulated kinases such as skMLCK and CaMKII that also contain the AID and CaMBR regions, albeit with variable CaM potency (e.g., 0.7, 1, and 40 nM for smMLCK,²⁹ skMLCK,³³ and CaMKII,³⁵ respectively). Another member of the MLCK family, striated muscle preferentially expressed protein kinase (SPEG), contains two kinase domains,³⁶ between which is a CaM binding region.³⁶ While it is demonstrated that SPEG does not require CaM for substrate phosphorylation,^37,38 the presence of the CaM binding region suggests a regulatory role of CaM on SPEG function. Similarly, another dual kinase protein within the MLCK family, obscurin, also has a CaM binding motif located in the interkinase region.³⁶

The motivation of this study is to provide insights into the molecular interactions that drive MLCK activation to spur future mechanistic studies. Toward this end, we propose a series of mutations at contact sites we identified (Figure 6C). For the gained contacts, mCaM K13 forms hydrogen bonds with the backbone of smMLCK P1160 and M1662. K13A/E mutants could therefore disrupt such hydrogen bonding interactions. smMLCK S1659 is in close proximity to mCaM K13, therefore, S1659D might lead to stronger interactions. Conversely, S1659K could disrupt the interactions. S17 is enclosed in a hydrophobic environment, hence S17L could potentially strengthen the hydrophobic interactions. In addition, the side chain of smMLCK M1662 is sandwiched by the aromatic side chains of mCaM F16 and F64. M1662F could form Pi–Pi stacking among the three aromatic side chains, leading to stronger interactions. smMLCK L1658 is in close proximity to mCaM S17. L1658E could form a hydrogen bond with S17. For the lost contacts, mCaM R37 forms a salt bridge with E1553. R37A/E could disrupt this interaction and potentially facilitate the removal of the AID. However, the resulted electrostatic repulsion of R37E might lead to distortion of the MLCK kinase domain similar to T34K. In summary, we propose the following mutants. Mutants that could enhance MLCK activation by CaM: mCaM S17L, smMLCK L1658E, S1659D, and M1662F by strengthening the interactions at the gained contact site, and mCaM R37A/E by weakening interactions at the lost contact site. Mutants that could impair MLCK activation by CaM: mCaM K13A/E and smMLCK S1659K by weakening the interactions at the gained contact site.

Our study provides rationale for further investigation of plant-based CaM homologues toward differentially regulating MLCK. We have already leveraged this CaM-engineering strategy in our efforts to counter cardiac arrhythmias,³⁹ by gene delivery of a CaM variant that modulate ryanodine receptor gating. In a similar manner, engineering and targeted delivery of CaM variants could treat diseases stemming from the malfunction or dysregulated expression of MLCK, such as pulmonary arterial hypertension.²

ABBREVIATIONS

CaM

calmodulin

sCaM-1

soybean CaM isoform 1

sCaM-4

soybean CaM isoform 4

smMLCK

smooth muscle myosin light chain kinase

skMLCK

skeletal muscle myosin light chain kinase

AID

autoinhibitory domain

CaMBR

CaM binding region

RLC

myosin regulatory light chain

CN

calcineurin

CaMKII

Ca²⁺/CaM-dependent protein kinase II

SPEG

striated muscle preferentially expressed protein kinase

SMC

smooth muscle cells

MD

molecular dynamics

PCA

principal component analysis

pLDDT

AlphaFold predicted local-distance differencetest

Data Availability Statement

Publicly available software, including visual molecular dynamics (VMD), PyMOL, and matplotlib library, was used for structure examination and figure preparation. ClusPro was used for protein–protein docking. Rosetta 3 was used for comparative modeling and loop modeling. AMBER18 was used for all MD simulations. AmberTools18 was used for system preparation (TLEAP) and analysis (CPPTRAJ). To facilitate data reproduction, Rosetta input files and simulation input files as well as all code written in support of this publication are publicly available at https://github.com/huskeypm/pkh-lab-analyses. Raw, full-length MD simulation data are also available by request and are redundantly stored on the Stritch School of Medicine computing center. Finalized simulation data are also available at 10.5281/zenodo.8075316 (Zenodo).