Seeking the exclusive binding region of phenylalkylamine derivatives on human T‐type calcium channels via homology modeling and molecular dynamics simulation approach

Abstract

Pharmaceutical features of phenylalkylamine derivatives (PAAs) binding to calcium channels have been studied extensively in the past decades. Only a few PAAs have the binding specificity on calcium channels, for example, NNC 55‐0396. Here, we created the homology models of human Ca_v3.2, Ca_v3.3 and use them as a receptor on the rigid docking tests. The nonspecific calcium channel blocker mibefradil showed inconsistent docking preference across four domains; however, NNC 55‐0396 had a unique binding pattern on domain II specifically. The subsequent molecular dynamics (MD) simulations identified that Ca_v3.1, Ca_v3.2, and Ca_v3.3 share domain II when Ca²⁺ appearing in the neighbor region of selective filters (SFs). Moreover, free‐energy perturbation analysis suggests single mutation of lysine at P‐loop domain III, or threonine at the P‐loop domain II largely reduced the total amount of hydration‐free energy in the system. All these findings suggest that P‐loop and segment six domain II in the T‐type calcium channels (TCCs) are crucial for attracting the PAAs with specificity as the antagonist.

Predicted Ca²⁺ interaction domains during 200 ns simulation period. For α₁H, Ca²⁺ ions are more likely to dock domain I (A) and II (B); however, for α₁I: Ca²⁺ ions are more likely to dock domain II (C) and IV (D).

​

Abbreviations

EEF

external electric force field

FEP

free‐energy perturbation

LCC

L‐type calcium channel

MD

molecular dynamics

NNC 55‐0396

(1S,2S)‐2‐(2‐(N‐[(3‐benzoimidazol‐2‐yl)propyl]‐N‐methylamino)ethyl)‐6‐fluoro 1,2,3,4‐tetrahydro‐1‐isopropyl‐2‐naphtyl cyclopropanecarboxylate dihydrochloride

PAAs

phenylalkylamine derivatives

PCA

principal component analysis

RMSD

root‐mean‐square‐displacement

SF

selective filter

SMD

steering molecular dynamics

TCCs

T‐type calcium channels

1. INTRODUCTION

T‐type calcium channels (TCCs) belong to the one sort of voltage‐dependent calcium channel family. They are known to be activated by membrane depolarization, conducting inward currents with a small single‐channel conductance. Roles of TCCs in controlling the hormone and neurotransmitter release under various conditions have been extensively studied in the past decades. ¹ , ² , ³ , ⁴ , ⁵ The distribution of TCCs can be found in pancreatic β‐cells, ⁶ heart, ⁷ and neuron cells. ⁸ In β‐cells, overexpressed TCC increased [Ca²⁺]_i followed by generating frequent random calcium spikes. ⁹ Humans with gradually raised calcium concentration will more likely develop type 2 diabetes in later life. ¹⁰ In cardiomyocytes, the TCC currents significantly affect the later stage of the action potential. ¹¹ In neuron cells, the development of chronic neuropathic pain due to spinal cord injury is contributed by the increased activity of TCCs. ¹²

When the inhibitor binds to the calcium channel, it cuts down the Ca²⁺ pathway by allosterically changing the pore conformation or physical blocks in the pore as a plug. ¹³ As the earliest launched TCC inhibitor, mibefradil was initially developed for blocking L‐type calcium channel (LCC) and showed the promising effect of blocking TCCs in vitro. ¹⁴ Such phenylalkylamine derivatives (PAAs) are more likely to behave as a physical plug when interacting with TCCs. Unfortunately, mibefradil was withdrawn from the market due to the interaction with other drugs due to its effect on P450. ¹⁵ Because phenylalkylamine‐based TCC antagonists are derived from LCC blockers, only some of them have specificity to block the TCC in vivo. ¹⁴ Recently discovered drug Z944 ¹⁶ showed an excellent specificity and potent to block the TCC for treating epilepsy ¹⁷ and neuronal pain. ¹⁸ In contrast to phenylalkylamines that physically block the pore, Z944 was reported to change the conformation of TCC by shifting the α to π helix at domain II and further shut down the calcium currents. ¹⁹

In this study, we created two homology models for human Ca _v 3.2 (UniProt id: O95180) and Ca _v 3.3 (UniProt id: Q9P0X4) in terms of the structure of human Ca _v 3.1 at apostate. A total of three TCC structures were used in subsequent molecular dynamics (MD) simulations. Mibefradil, NNC‐55‐0396 were selected as the representative ligands in the MD to mimic the real cases when TCCs interact with different sorts of an antagonist. Also, the external electric force field MD and steering MD simulation was utilized to find the pathway of Ca²⁺ permeation when it penetrates the channel pore with and without introducing the blockers.

2. MATERIALS AND METHODS

2.1. Protein comparative modeling

Two human TCC 3D structures (α₁H and α₁I) were created by using the comparative modeling package from Rosetta. ²⁰ The Cryo‐EM structure of human Ca_v3.1(PDB id: 6KZO) at the apostate was selected as input for Rosetta to create the homology model of α₁H and α₁I. The energy‐based clustering method ²¹ was applied to categorize the predicted models before the quality evaluation process. To filter out the low quality of structure, PROCHECK ²² and WHATCHECK ²³ were employed on clustered data. Finally, the structures that hold the lowest Rosetta energy score will be selected as targets for the next tests, that is, rigid docking and MD simulations.

2.2. Rigid docking with selected PAAs

The 3D structure of mibefradil and NNC 55‐0396 was downloaded from the PubChem online database. We used Open Babel ²⁴ to convert the compound format and Frog ²⁵ to find the coordinate of compound conformers in 3D space. To find out the possible binding sites between existing antagonists and TCCs, we used AutoDock Vina ²⁶ to simulate the rigid docking process. The searching box was set at the center of the SF with the grid spacing 0.375 Å and 40 grid points along X, Y, and Z directions. The number of predict binding modes is set as 10, and the random seed number was set as −1460306363.

2.3. MD simulations

2.3.1. Ion channel membrane assembly

The CHARMM‐GUI Membrane Builder ²⁷ was used to build the membrane system. The missing residues were modeled by GalxyFill. The heterogeneous lipid bilayer was created by choosing phosphatidylcholine lipid and phosphatidylethanolamine lipid ratio as 2 over 1 on both inner and outer leaflets of the membrane. The water thickness was set as 22.5 Å. The system size along the X and Y dimensions is set as 120 Å. About 150 mM calcium chloride solution was added into the 3D rectangle computational domain. To generate the parameter and force field file for selected TCC blockers, the Ligand Reader and Modeler ²⁸ was used. Overall, the system was formed as an isothermal‐isobaric ensemble at temperature 310 K.

2.3.2. Classic MD simulation

The force field to support all MD simulations was set as CHARMM36m. All MD productions were conducted on the precompiled NAMD‐2.14b GPU‐acceleration Linux version. Since MD cases prepared by CHARMM‐GUI Membrane Builder need to be equilibrated before long‐term standard running. The energy minimization process was performed under gradually reduced restrain forces over six‐constitute steps. To keep the system stable, the integration time step was set as 2 femtoseconds (fs) for the first and second steps, then adjust to 1 fs for the others. As long‐term standard MD production, the integration time step was set as 2 fs and the position and velocity of all atoms in the system will be recorded every 10 picoseconds (ps). The maximum periodic electrostatics calculation was based on the 1 Å grid size with periodic boundary conditions. To calculate the nonbound interactions, a cutoff of 12 Å is used with switch distance 10 Å as input conditions for solving Langevin equations at temperature 310 K and 1 atm. For long‐term production, the trajectory information was collected over 100 nanoseconds (ns) timescale with zero restrictions added to it.

2.3.3. Simulation under the external electric force field

The Cryo‐EM structure of human α₁G at the apo statue shows the opening channel pore without ligand binding. To simulate the open channel process, an external voltage was required to apply to the lipid membrane. The external electric potential E_z along L_z can be defined by the equation: $E_z=\frac{V}{L_z\times 43.17}$, where V = −60 mV ²⁹ , ³⁰ and L_z is the average length of TCCs in Z direction at the last 10 ns from standard MD simulation. The numerical value of 43.17 was the force conversion coefficient used by NAMD‐2.14b. Total MD simulation time under −60 mV for all TCCs was set as 100 ns.

2.3.4. Structure identification through RMSD

For all the MD simulation cases, the structure clustering analysis was conducted when the overall system reaches equilibrium, typically, in the last 50 ns. The optimized number of clusters for a given length of MD data was determined by a comparison of the clustering results from various methods. ³¹

2.3.5. Steering MD

To simulate the Ca²⁺ influx through the channel pore, the steering molecular dynamics (SMD) method was employed. The clustering analysis of external electrical force field simulation provided the three most typical conformations for running SMD. We picked up one Ca²⁺ which most close to the SF as a pulling target. Then we applied a constant pulling velocity: V = −4e−5 Å/fs with spring constant K = 4 kcal/mol Å along z direction for 3 ns simulation. To minimize the artificial impact from fixed atoms in SMD simulation, we fixed the C‐alpha atoms at four geometrical symmetrical amino acids that away from channel pore, in which Pro¹¹⁶, Pro⁸⁴⁹, Gln¹³⁷³, and Val¹⁷⁰⁴ for α₁G; Pro¹¹⁶, Pro⁸⁰¹, Pro¹³¹⁶, and Thr¹⁶³⁵ for α₁H; Pro¹¹⁶, Pro⁶⁶⁹, Pro¹²¹³, and Thr¹⁵²⁶ for α₁I, respectively. With Tcl scripts to define all necessary parameters, the repeating rate of each structure was set as 20.

2.3.6. Finding calcium ion interaction pathway

The trajectory files generated by NAMD over the entire time‐span were analyzed in PyContact. ³² To find the interactions between Ca²⁺ and select molecules, the maximal interatomic distance for contact scoring was set as 5 Å, the cutoff angle for hydrogen bonds was set as 120°, and maximal distance between the hydrogen bond receptor and the H‐atom was set as 2.5 Å. Accumulate score was used to evaluate the interaction strength between Ca²⁺ and selected molecular.

2.3.7. Free energy perturbation analysis

The alchemical free‐energy perturbation (FEP) method ³³ was used to estimate the influence of Gibbs‐free energy by mutation of pre‐identified binding associated amino acids. ¹⁹ K^3p1462 and T^2p921 at α₁G, K^3p1405 and T^2p784 at α₁H, and K^3p1302 and T^2p742 at α₁I are mutated into alanine. Two FEP simulations were conducted for molecular in vacuum or immersed in bulk water. In every simulation, 20 equally stratified external parameters λ ³⁴ in the range [0, 1] were used to sample the energy variation in 50 ps with coupled intramolecular interactions. The net solvation free energy change was computed as follows: $\mathrm{\Delta }\mathrm{\Delta }{G}_{\text{mutation}}=\mathrm{\Delta }{G}_{\text{hydr}}^2‐\mathrm{\Delta }{G}_{\text{hydr}}^1$, where $\mathrm{\Delta }{G}_{\text{hydr}}^2$ is hydration energy in the solvation state and $\mathrm{\Delta }{G}_{\text{hydr}}^1$ is the hydration energy at the isolate state. The results of FEP were processed by ParseFEP in VMD 1.9.3 at temperature 310 K, Gram–Charlier order at 3, with Gaussian approximation and BAR estimator. ³⁵

2.4. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY, ³⁶ and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20. ³⁷

3. RESULTS

3.1. Homology modeling

Compare to other calcium channel templates, 6KZO returns the standard protein BLAST identity value as 62.58%, 57.02% for the full length of amino acid (AA) sequences (Table 1). We only considered amino acids aligned with 139 to 1906 to generate the homology model for α₁H and α₁I to reduce the computational cost. The detailed alignment of amino acids for three TCC can be found in the supporting information (Figure S1). Based on the target protein AA sequences, the energy‐based clustering method has identified 10 and 19 featured structures for α₁H and α₁I, respectively. The Rosetta Comparative Modeling method uses the continuous alignment results of α₁H and α₁I to α₁G for generating the homology models. The AAs that are not included in 6KZO will affect the quality of predicted protein structures. We found that after trimmed these regions out, the highest Ramachandran Z‐score can go to 99.5% (Figure 1A1), and side‐chain planarity can reach 0.42 for selected α₁H homology modeling result (Figure 1A2). Meanwhile, the highest Z‐score and side‐chain planarity is 99.7% (Figure 1B1) and 0.425 for trimmed α₁I homology modeling results. The full structure comparison before and after trimmed for the unnecessary region of α₁H and α₁I is summarized in Table S1.

TABLE 1.

Protein basic local alignment search tool (BLAST) results on the identity score of a given length of amino acids between template proteins and targets.

	H‐Cav3.1 (full length)	R‐Cav1.1 (full length)	Bac‐CavAb (full length)
H‐Cav3.2	62.58%	31.47%	23.93%
H‐Cav3.3	57.02%	31.09%	24.46%

Abbreviations: Bac‐Ca_vAb (5KMH), bacterial Ca_vAb; H‐Ca_v3.1, human Ca_v3.1; R‐Ca_v1.1, rabbit Ca_v1.1 (3JBR).

FIGURE 1.

The distribution of torsional angles phi (ϕ) and psi (ψ) indicates over 99% of residues following in the favored region. (A1): the Ramachandran plot of selected α₁H model and (B1): the Ramachandran plot of selected α₁I model. The homology modeling of α₁H (A2) and α₁I (B2) structure show structural similarity to the template α₁G. The nonpolarized side‐chain of SFs from the domain I to IV are colored as cyan, blue, green, and red.

3.2. Rigid docking with selected PAAs

Vina virtual screening process finds a similar binding pattern between receptor and ligands. The hydroxy‐group of S^4p1776 domain IV at α₁G forms hydrogen bonds with a hydrogen atom at ammonia (N₃) on the imidazole ring in the benzimidazole moiety of mibefradil. One hydrogen atom from the central ammonium of NNC 55‐0396 could also form a hydrogen bond with carbonyl‐oxygen in the amide of N^2p952 domain II at α₁G (Figure 2A); For α₁H, one oxygen from carboxyl groups on the side‐chain of mibefradil forms a hydrogen bond within 3.5 Å at S^4p1707 from domain IV. Ammonia (N₃) on the imidazole ring in the benzimidazole moiety of NNC 55‐0396 generates a connection with hydrogen atom from K^3p1405 from domain III (Figure 2B); For α₁I, one hydrogen atom from ammonia (N₃) on the imidazole ring in the benzimidazole moiety of NNC 55‐0396 could bind to the lipophilic side‐chain of L^2p776 at domain II. In comparison, the mibefradil is less likely to place at the center of the channel pore after binding to the amino acid (Figure 2C). The predicted binding affinities between testing drugs and TCCs from 2.67 to 25.99 mM (Figure 2D) are matched to other published studies. ¹⁴ , ³⁸ , ³⁹ , ⁴⁰

FIGURE 2.

Comparison of amino acid binding sites between mibefradil and NNC 55‐0396 across three T‐type calcium channel (TCC) structures. Cyan: domain I; green: domain II; yellow: domain III; and red: domain IV. Sidechains of four key amino acids are given with partially displayed domains for better visualization. (A) mibefradil (magenta) binds to S^4p1776 and NNC 55‐0396 (wheat) binds to N^2p952 at α₁G; (B) mibefradil (marine) binds to S^4p1707 and NNC 55‐0396 (yellow) binds to K^3p1405 at α₁H; (C) mibefradil (green) binds to N^1s314 and NNC 55‐0396 (violet) binds to L^2p776 at α₁I. (D) predict the binding affinity of mibefradil and NNC 55‐0396 varied across α₁G (red), α₁H (blue), and α₁I (black) (unit: kcal/mol).

3.3. MD simulations and analysis

The time‐dependent root‐mean‐square‐displacement (RMSD) plot for C‐alpha and whole protein indicates that all three TCC membrane protein complex reaches the equilibrium at 20 ns (Figure 3A–α₁G, Figure S2A–α₁H, and Figure S3A–α₁I). Employing the first‐five ranking eigenvalue, ⁴¹ the structural variation and conformation changes among α₁G, α₁H, and α₁I are 55.6% (Figure 4B), 61.6% (Figure S2B), and 56.5% (Figure S3B), respectively, while the proportion of variance decreasing monotonically with increased eigenvectors. For the first‐three ranking eigenvalue, 50.9% α₁G structure was covered. And this number increased to 58.2% on α₁H and 51.4% on α₁I. Conducting principal component analysis (PCA) with three eigenvectors, we did not observe significant conformation changes in the transmembrane part at the last 50 ns simulation for all three TCC structures. The local structural differences among each cluster can be quickly distinguished by checking the shape of the transmembrane region between segment 5 and segment 6 (Figure 3B, clusters 1–3; Figure S2B, clusters 1–3; Figure S3B, clusters 1–3). For α₁G, expect for P‐loop domain I which dynamic behavior is represented by high RMSD value, the rest of the domains have almost identical structure variation patterns in both sorted and unsorted trajectory data within 50 ns (Figure S4A); There is no clear structure variation of P‐loop domains I and IV for α₁H till the last 5 ns (Figure S4A); For α₁I, P‐loop domain IV has more unstable local fluctuations than other domains (Figure S4C).

FIGURE 3.

Structure variation of a₁G during the standard molecular dynamics (MD) simulation. (A) Comparison of the variation of root‐mean‐square‐displacement (RMSD) over time between C‐alpha and whole protein during 100 ns; (B) principal component analysis (PCA) analysis on MD data from 50 to 100 ns.

FIGURE 4.

Combine the standard molecular dynamics (MD) simulation (0–100 ns) and external electric force field (EEF) MD simulation (100–200 ns) results show two extracellular Ca²⁺ ions that are attracted by two selected filters. (A) Ca³⁷ binds to E^1p354 at domain I since t = 30 ns (α₁G); (B) Ca¹⁵ binds to E^2p923 at domain II at t = 8 ns (α₁G); (C) Ca⁵⁸ binds to E^1p280 at domain I (α₁H); (D) Second Ca³⁸ binds to E^2p876 at domain II (α₁H); (E) First Ca³¹ binds to E^2p744 at domain II (α₁I); (F) Second Ca⁵⁶ binds to E^4p1601 at domain IV (α₁I).

The classic MD simulation (0–100 ns) with the following EEF MD (100–200 ns) simulation did not detect the penetration of calcium ion through the membrane in α₁G, α₁H, and α₁I. For α₁G, Ca³⁷ binds to E^1p354 domain I at t = 30 ns (Figure 4A), and Ca¹⁵ binds to E^2p923 domain II at t = 8 ns (Figure 4B). Before simulation reaches steady‐state, Ca³⁷ is initially shared by two SFs at domains II and III for 25 ns, then it jumps to E^1p354 domain I (Figure S5A and B). SF at domain IV has less attraction to calcium ions. For α₁H, Ca⁵⁸ binds to E^1p280 domain I at t = 0 (Figure 4C) and Ca³⁸ binds to E^2p876 domain II after 25 ns (Figure 4D). SF D^3p1406 at domain III and D^4p1710 at domain IV show limited binding affinity to Ca²⁺ (Figure S5C and D). For α₁I, Ca³¹ binds to E^2p744 at domain II (Figure 4E) with unstable binding behavior and Ca⁵⁶ firmly binds to E^4p1601 domain IV (Figure 4F) after 25 ns. The SF at domains I and III shows fewer interests in binding Ca²⁺ (Figure S5E and F). At the end of EEF MD simulation, α₁G and α₁H have two calcium ions bind to the domaind I and II (Figure 5A and B); however, two calcium ions go to domains II and IV at α₁I (Figure 5C).

FIGURE 5.

Interaction between Ca²⁺ and amino acids in the channel pore. The nonpolar side‐chain of four SFs from domains I—IV are given as color order: cyan (domain I), green (domain II), yellow (domain III), and red (domain IV); (A and B) Ca²⁺ (blue dot) contacts to SF domain I and another Ca²⁺ (hotpink) contacts to SF domain II for α₁G and α₁H at the end of external electric force field (EEF) simulation, respectively; (C) Ca²⁺ (blue dot) contacts to SF domain II and another Ca²⁺ (hotpink) contacts to SF domain IV for α₁I at the end of EEF simulation; (D) Ca²⁺ interacts with T^2p921, Q^2p922, and D^2p924 from α₁G domain II at t = 0.2 ns in steering molecular dynamics (SMD) simulation; (E) Ca²⁺ interacts with T^2p874, I^2p872, and Q^2p875 from α₁H domain II at t = 0.8 ns in SMD simulation; (F) Ca²⁺ interacts with T^2p742, Q^2p743 from α₁I domain II, and K^3p1302 from α₁I domain II at t = 0.5 ns in SMD simulation.

Charged amino acids or amino acids with polarity potential are more likely to interact with the Ca²⁺ when the interatomic distance is smaller than 10 Å. Our SMD tests find T^2p921, Q^2p922, and D^2p924 at domain II from α₁G; T^2p874, I^2p872, and Q^2p875 at domain II from α₁H; T^2p742, Q^2p743 at domain II; and K^3p1302 at domain III from α₁I can interact with pulling Ca²⁺ due to the short atom–atom distance (Figure 5D–F). During the 3 ns SMD simulation, a maximum pulling force is less than −15 pN along the z‐axis for all three TCCs structures. The reaction coordinate in the tests is fixed in the negative direction along the z‐axis; three clusters have different initial z‐coordinate of Ca²⁺, which yield the varied initial force profile. For all three TCC structures, throughout the Ca²⁺ penetration, the absolute value of interatomic force first decreases in the SF region, the increase when Ca²⁺ wants to escape from the SFs. Eventually, it reaches the maximum before leaving the intercellular gate. Expect for cluster 1s at α₁H and α₁I, an arch‐shape force was generated when Ca²⁺ moving towards the negative direction of Z (Figure 6).

FIGURE 6.

Nonlinear increased pulling force (blue) in response to moving one Ca²⁺ across three channels, (A) α₁G, (B) α₁H, and (C) α₁I. The error bar is colored by gray for α₁G and green for selected cluster 1 structure, yellow for selected cluster 2 structure, and magenta for selected cluster 3 structure from last 50 ns of external electric force field (EEF) molecular dynamics (MD) simulation trajectory of α₁H, and α₁I.

Replace K^3p1462 in α₁G with Phe or Gly will shift the activation and decrease the drug sensitivity. ¹⁹ FEP analysis finds in water a mutation caused total Gibbs‐free energy to increase by 43.83, 40.52, and 36.9 kcal/mol on α₁G, α₁H, and α₁I, respectively. In vacuum, these numbers become 73.96, 43.58, and 62.27 kcal/mol on α₁G, α₁H, and α₁I, respectively. Thus, the contribution of mutation lysine (K^3p1462, K^3p1405, and K^3p1302) at P‐loop domain III on whole protein hydration‐free energy ΔΔG_(K‐A) yields: −30.13, −3.06, and −25.37 kcal/mol. FEP analysis on T^2p921, T^2p874, and T^2p742 also finds the polarity of threonine in response to the mutation test. In water, ΔG variation is 13.08, −25.47, and 14.65 kcal/mol for α₁G, α₁H, and α₁I, respectively. These numbers become 13.8, 11.46, and 14.54 kcal/mol (Table 2). Details of how Gibbs‐free energy in response to selection of λ during the alchemical reaction in water or vacuum are given in Figure S6. The quality control of FEP analysis is given in Figure S7.

TABLE 2.

Free Gibbs energy (ΔG) and free hydration energy (ΔΔG) variation introduced by mutation lysine/threonine into alanine (unit: kcal/mol)​.

	K1462/K1405/K1302		ΔΔG	T921/T874/T742		ΔΔG
	ΔG(Water)	ΔG(Vacuum)		ΔG(Water)	ΔG(Vacuum)
α1G	43.83	73.96	−30.13	13.08	13.8	−0.72
α1H	40.52	43.58	−3.06	−25.47	11.46	−36.93
α1I	36.9	62.27	−25.37	14.65	14.54	0.11

4. DISCUSSION

Emerging evidence points out the pathological role of TCCs that is associated with the progression of different diseases. Previous works suggest that nitrile and isopropyl groups in phenylalkylamine serve a role to guide the drug to the position of Ca²⁺; this function remains positive if the nitrile is replaced with other high electronegative potential elements such as oxygen or sulfur. ⁴² The recently deposited crystallization structure of human Ca_v3.1 offers a great opportunity to explore the Ca²⁺ binding mechanism on all TCCs. ¹⁹

It is suggested that at least 20,000 decoys should be generated if we go through the de novo homology modeling process. ⁴³ Using the human Ca_v3.1 structure as a template, we set the sampling number at 15,000 for each comparative modeling case. Although the homology modeling template is constructed based on the splice variant containing a deletion of amino acids within the I–II liker, there are not AAs breaks from SF to the intercellular gates. Therefore, delete the regions that are not aligned with α₁G from selected clustering results should not affect the docking preference of Ca²⁺ to α₁H and α₁I.

Vina does predict binding free energy between ligand and receptor less accurate than MM‐PBSA. ⁴⁴ However, it gives a direct visualization for filtering out the incorrect binding poses between ligand and receptor. Using the predefined homology models, we confirmed that the mibefradil and NNC 55‐0396 have different binding regions on the same structure but consistent binding regions from α₁H to α₁I. This finding partially explained why NNC 55‐0396 is an exclusive TCCs blocker. ¹⁴ The existence of positively charged lysine at P‐loop domain III will largely reduce the probability of Ca²⁺ binds to SF at domain III. Our MD simulations further confirmed that SF at P‐loop domain II is the common region to dock Ca²⁺ across all TCCs structure. Two AAs (K^3p1462/T^2p921) that affect the binding affinity of z944 on α₁G ¹⁹ also show a clear variation of free hydration energy if mutate to alanine in water. Additional tests results for free hydration energy variation due to the mutation of calcium‐binding associated amino acids are available upon request. We used water as solvation in FEP analysis because only K^3p1462 has a direct impact on α₁G current and conductance. More sophisticated conditions should add into solvate for calculating the influence of polarized AA if it has the potential to be the target site to bind.

DISCLOSURE

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

AUTHOR CONTRIBUTIONS

YL, GYL, and ZC: Design, programming, and writing; KZ, A.E., ML, NZ: Supervision.

ETHICS

N/A.

Open Research Badges

This article has earned Open Data, Open Materials and Preregistered Research Design badges. Data, materials and the preregistered design and analysis plan are available in the article.

Supporting information

Figure S1‐S7‐Table S1

DATA AVAILABILITY STATEMENT

Human Ca_v3.1(6KZO), rabbit Ca _v 1.1 (3JBR), and bacterial Ca _v Ab (5KMH) Cryo‐EM structures were downloaded from RCSB PDB. Amino acid sequences of Ca_v3.2 (O95180) and Ca_v3.3 (Q9P0X4) were downloaded from UniProt. The 2D structures of mibefradial and NNC 55‐0396 were downloaded from PubChem. The homology modeling and MD simulation data are available on request from the authors.