Molecular dynamics study reveals key disruptors of MEIG1–PACRG interaction

Abstract

Interactions between the meiosis-expressed gene 1 (MEIG1) and Parkin co-regulated gene (PACRG) protein are critical in the formation of mature sperm cells. Targeting either MEIG1 or PACRG protein could be a contraceptive strategy. The W50A and Y68A mutations on MEIG1 are known to interrupt the MEIG1–PACRG interactions resulting in defective sperm cells. However, the details about how the mutants disrupt the protein–protein binding are not clear. In this study, we reveal insights on MEIG1 and PACRG protein dynamics by applying Gaussian-accelerated molecular dynamics (GaMD) simulations and post-GaMD analysis. Our results show that the mutations destabilize the protein–protein interfacial interaction. The effect of the Y68A mutation is more significant than W50A as Y68 forms stronger polar interactions with PACRG. Because both human and mouse models demonstrate similar dynamic properties, the findings from mouse proteins can be applied to the human system. Moreover, we report a potential ligand binding pocket on the MEIG1 and PACRG interaction surface that could be a target for future drug design to inhibit the MEIG1–PACRG interaction. PACRG shows more qualified pockets along the protein–protein interface, implying that it is a better target than MEIG1. Our work provides a fundamental understanding of MEIG1 and PACRG protein dynamics, paving the way for drug discovery in male-based contraception.

1 |. INTRODUCTION

Meiosis-expressed gene 1 (MEIG1) is a protein that plays an essential role in the formation of functional sperm cells in the spermiogenesis process.^1–3 Spermiogenesis, the final stage of spermatogenesis, is the maturation of spermatids to spermatozoa characterized by the elongation and formation of flagellum.^4,5 In elongating spermatids, MEIG1 binds to the Parkin co-regulated gene (PACRG) protein to form a temporary complex in the manchette structure that assists with the elongation process by transporting cargo proteins for sperm formation.⁶ The previous study has shown that global disruption of the Meig1 gene in mice results in severely defective spermiogenesis and failure of flagellogenesis,⁷ which sheds light on an alternative strategy for male-based contraception development.⁸ By targeting the formation of the MEIG1–PACRG complex structure, one can expect to inhibit the function of the complex in sperm formation, which will then lead to failed sperm production.

The structure of the mouse MEIG1 was solved in 2016.⁷ The human structure of the MEIG1–PACRG complex was released recently.⁹ MEIG1 adopts a unique fold with alternating α-helices and β-sheets (Figure 1). The MEIG1 structure also reveals a patch of residues that have been identified as potential mediators of protein–protein interactions for the binding of PACRG. The binding affinity between MEIG1 and PACRG has been experimentally determined. The results show that the W50A, K57E, F66A, and Y68A mutations on MEIG1 substantially reduce the binding of the two proteins.⁷ These four mutated residues are exposed on the surface that interfaces with PACRG in the protein–protein interaction. Particularly, the single mutation of Y68A shows the most profound effect on the reduced binding and the interruption of the cargo transport system necessary for sperm formation.³ Furthermore, compared with the single mutations, a double mutation of W50A and Y68A further reduced the MEIG1–PACRG affinity. However, it is not clear how the mutations affect the protein–protein interaction and why Y68A has a stronger effect than W50A. By addressing these questions, we could design potential compounds that may inhibit the MEIG1–PACRG interaction as a new male-based contraceptive.

FIGURE 1.

Structure of meiosis-expressed gene 1 (MEIG1, magenta) in complex with Parkin co-regulated gene (PACRG, green) with Y68 and W50 shown in the bond form.

Molecular dynamics (MD) simulations have been broadly applied in uncovering the physical movement of a target protein in solution.^10–12 Gaussian-accelerated MD (GaMD) is an MD-based enhanced sampling technique.^13,14 By adding a harmonic boost to the original potential surface, GaMD can reduce the energy barrier and provide unconstrained sampling that is advantageous for simulating long-timescale conformational transitions of biomolecules within a short simulation time. GaMD has been applied successfully in many situations to simulate protein dynamics,^13,15,16 including protein folding,^17–19 ligand binding,^17–24 protein–protein interactions,^25–28 membrane protein dynamics,^17,20,29 protein–nucleic acid interactions,³⁰ protein–carbohydrate interactions,³¹ and ion channel motions.³² As GaMD is a powerful technique in conformational sampling, it serves as a primary tool to simulate the motion of MEIG1 and PACRG proteins here.

In this study, we aim to reveal how the single mutations W50A and Y68A on MEIG1 alter the local and global protein dynamics and disrupt the MEIG1–PACRG interaction. To approach this goal, we performed simulations on both wild-type and mutated forms. We also studied multiple protein models obtained from the Protein Data Bank (PDB) and homology modeling, including apo-PACRG, apo-MEIG1, and holo-MEIG1 in complex with PACRG, which will allow us to address the protein dynamic changes before and after binding. To understand the difference between human and mouse proteins, systems from both human and mouse models are included in this work. The analysis of protein flexibility, hydrogen bonds, and energetic changes will quantitatively describe how the mutations affect protein motion. Finally, by computational modeling, we reveal potential ligand binding pockets on both MEIG1 and PACRG protein surfaces to help with future binder design in inhibiting MEIG1–PACRG interactions.

2 |. MATERIALS AND METHODS

2.1 |. Molecular models

In this work, 27 models were created for simulations and analysis (Table 1). Three protein models were obtained from PDB, mouse apo-MEIG1 (PDB code: 2N2Y)⁷ and two human MEIG1 in complex with PACRG (PDB code: 6NDU and 6UCC).⁹ 2N2Y contains 20 conformations, and two conformations (Nos. 1 and 11) were selected. The two human complexes share similar structural features; however, different ligands and ions were shown in the structures for crystallization. As all ligands and ions would not influence protein dynamics, they were removed in simulations while water molecules were kept. The mouse apo-PACRG was constructed from the amino acid sequence using I-TASSER³³ protein structure predictor. The mouse MEIG1–PACRG complex was created by aligning mouse apo-MEIG1 and apo-PACRG with the human MEIG1–PACRG complexes using the visual molecular dynamics (VMD)³⁴ program. Human apo-MEIG1 and apo-PACRG models were created by removing either PACRG or MEIG1 from 6NDU and 6UCC. Mutant models were built by manually substituting MEIG1 residues W50 and Y68 with alanine on the above models. MultiSeq³⁵ tool in VMD was used to calculate sequence identity between the human and mouse proteins.

TABLE 1.

Molecular models.

Model No.	PDB ID	Organism	Protein	Mutation
1	6NDU	Human	MEIG1–PACRG	Wild type
2	6UCC	Human	MEIG1–PACRG	Wild type
3	6NDU	Human	MEIG1–PACRG	W50A
4	6UCC	Human	MEIG1–PACRG	W50A
5	6NDU	Human	MEIG1–PACRG	Y68A
6	6UCC	Human	MEIG1–PACRG	Y68A
7	6NDU	Human	MEIG1	Wild type
8	6UCC	Human	MEIG1	Wild type
9	6NDU	Human	MEIG1	W50A
10	6UCC	Human	MEIG1	W50A
11	6NDU	Human	MEIG1	Y68A
12	6UCC	Human	MEIG1	Y68A
13	6NDU	Human	PACRG	Wild type
14	6UCC	Human	PACRG	Wild type
15	2N2Y-1 and I-TASSER	Mouse	MEIG1–PACRG	Wild type
16	2N2Y-11 and I-TASSER	Mouse	MEIG1–PACRG	Wild type
17	2N2Y-1 and I-TASSER	Mouse	MEIG1–PACRG	W50A
18	2N2Y-11 and I-TASSER	Mouse	MEIG1–PACRG	W50A
19	2N2Y-1 and I-TASSER	Mouse	MEIG1–PACRG	Y68A
20	2N2Y-11 and I-TASSER	Mouse	MEIG1–PACRG	Y68A
21	2N2Y-1	Mouse	MEIG1	Wild type
22	2N2Y-11	Mouse	MEIG1	Wild type
23	2N2Y-1	Mouse	MEIG1	W50A
24	2N2Y-11	Mouse	MEIG1	W50A
25	2N2Y-1	Mouse	MEIG1	Y68A
26	2N2Y-11	Mouse	MEIG1	Y68A
27	I-TASSER	Mouse	PACRG	Wild type

Note: The models extracted from PDB are underlined. The mouse model 2N2Y⁷ was obtained from solution NMR, which contains 20 conformations. The two conformations (Nos. 1 and 11) were chosen for GaMD simulations. Models 15, 17, 19, 21, 23, and 25 were from the first NMR conformation; models 16, 18, 20, 22, 24, and 26 were from the 11^th NMR conformation. The mouse PACRG model was predicted using the I-TASSER server.³³

Abbreviations: GaMD, Gaussian-accelerated molecular dynamics; MEIG1, meiosis-expressed gene 1; PACRG, Parkin co-regulated gene; PDB, Protein Data Bank.

2.2 |. MD simulation protocol

For each model system listed in Table 1, we first performed the PROPKA program to determine the protonation state of the protein sidechains using the PDB2PQR server,³⁶ from which histidine residues were assigned either a neutral charge, with a single proton on either the epsilon or delta nitrogen, or a positive charge with both the epsilon and delta nitrogen protonated. Mutations were substituted where desired. All MD-based simulations were performed using the AMBER18 package.³⁷ First, the protein models were minimized in three steps, including 500 steps of hydrogen minimization, 5000 steps of hydrogen and sidechain minimization, and 5000 steps of whole protein minimization. Next, the proteins were solvated using the TIP3P water model extending 12 Å from the surface while ions were added to neutralize the charges of the system. The system was minimized again by first minimizing the water molecules with 1000 steps, then minimizing the entire system with 5000 steps. A 12 Å cutoff was applied in the minimization, and the particle mesh Ewald summation³⁸ was turned on. After the multiple-step minimization, we equilibrated the system by gradually heating it from 50 to 300 K in steps of 50 K. Each equilibration simulation was 10 ps, except the 300 K equilibration was 100 ps. Finally, a 20 ns conventional MD simulation was performed at 300 K to ensure that the system converged to the appropriate thermodynamic internal energy. All MD simulations were performed using Langevin dynamics³⁹ with a collision frequency of 5 ps⁻¹ in an isothermal-isobaric (NPT) ensemble. Bonds containing hydrogen atoms were restrained using the SHAKE algorithm.⁴⁰ The time step of the simulation was set to 2 fs.

2.3 |. Gaussian-accelerated MD

GaMD is a robust computational method that allows users to obtain biomolecule conformations by unconstrained enhanced sampling.^13,14 It works by adding a harmonic boost potential to smooth the original potential energy surface and reduce the system energy barriers. When the system potential $V\left(\overrightarrow{r}\right)$ is lower than the reference energy $E$, the modified potential $V^{\mathrm{*}}\left(\overrightarrow{r}\right)$ of the system can be calculated as follows:

$$\begin{gathered} V^{\mathrm{*}}\left(\overrightarrow{r}\right)=V\left(\overrightarrow{r}\right)+\mathrm{\Delta }V\left(\overrightarrow{r}\right) \\ \mathrm{\Delta }V\left(\overrightarrow{r}\right)=\left\{\begin{array}{ll}\frac{1}{2}k(E-V(\overrightarrow{r}){)}^2,& \text{if}V\left(\overrightarrow{r}\right)<E\\ 0,& \text{if}V\left(\overrightarrow{r}\right)\ge E\end{array}\right. \end{gathered}$$ (1)

where $k$ is the harmonic force constant. All GaMD simulations include a 50 ns GaMD preparation run followed by a 200 ns GaMD production run. The GaMD preparation run begins with a short 2 ns of conventional MD performed at 300 K to collect the maximum, minimum, average, and standard deviation of the potential energy, used to calculate the harmonic force constant. Next, the boost potential is applied without updating parameters for 1 ns GaMD simulation. Last, for the GaMD preparation, the boost parameters are updated and used to calculate a new boost potential every 500 ps over a 50 ns GaMD simulation. For the GaMD production run, the boost potential is applied, and the boost parameters are fixed over the entire 200 ns simulation. For all GaMD simulations, a dual boost on both the dihedral and total potential energy was applied to enable maximum acceleration. The simulations were set to the lower bound energy limit. The upper limit of the standard deviation of the boost potential, ${\sigma }_0$, was set to 6.0 kcal/mol for both the dihedral and total potential energy terms. The resulting trajectories were collected every 10 ps for analysis.

2.4 |. Post-GaMD analysis

The VMD³⁴ program was used to visualize the GaMD trajectories and computed atomic distances and hydrogen bond occupancy. Atomic distances were measured from W50CB to K57CB, Y69O, and K71CB of MEIG1, and H137CB of PACRG; and from Y68CB to K58CB and F66CB of MEIG1, and W96CB, I100CB, and E101CD of PACRG (Figure 2C). The CPPTRAJ⁴¹ program in the AMBER package³⁷ was used to calculate root mean square deviation (RMSD), root mean square fluctuation (RMSF), correlation analysis, and pairwise nonbonded interaction energy. Nonbonded interaction energies were calculated every 10 ps between key MEIG1 residues W50, K57, K58, F66, Y68, Y69, and K71 and PACRG residues W96, I100, E101, and H137; and then averaged over the entire trajectory to get the pairwise energy. The difference between the pairwise energy of wild types and mutants is reported as:

$$\mathrm{\Delta }⟨E⟩=⟨E_{\text{wild}\text{type}}⟩-⟨E_{\text{mutant}}⟩$$ (2)

where $⟨E_{\text{wild}\text{type}}⟩$ and $⟨E_{\text{mutant}}⟩$ is the average pairwise energy of the wild type and mutant GaMD simulation trajectory, respectively.

FIGURE 2.

Distance change of W50, Y68, and nearby residues. (A) W50 is shown to interact with K57, Y69, and K71 of meiosis-expressed gene 1 (MEIG1) and with H137 of PACRG. (B) Y68 is shown to interact with K58 and F66 of MEIG1 and with W96, I100, and E101 of Parkin co-regulated gene (PACRG). The carbon atoms in MEIG1 and PACRG are colored as magenta and green, respectively. Red and blue represent oxygen and nitrogen atoms, respectively. (C) The distance of W50–K57, W50–Y69, W50–K71, W50–H137, Y68–K58, Y68–F66, Y68–W96, Y68–I100, and Y68–E101 was computed along the simulation time.

Molecular Mechanics Poisson–Boltzmann Surface Area (MMPBSA) and Molecular Mechanics Generalized–Born Surface Area (MMGBSA)⁴² were used to calculate the interaction energy between MEIG1 and PACRG. The interaction binding energy, $\mathrm{\Delta }{E}_{\text{bind}}$, associated with the binding of MEIG1 to PACRG to form a complex is calculated as follows:

$$\mathrm{\Delta }{E}_{\text{bind}}=⟨E_{\text{complex}}⟩-⟨E_{\text{MEIG}1}⟩-⟨E_{\text{PACRG}}⟩$$ (3)

where the bracket $⟨E⟩$ denotes the average energy calculated using the topology of either the MEIG1–PACRG complex, apo-MEIG1, or apo-PACRG and computed along the single trajectory obtained from the MEIG1–PACRG complex simulation. The changes in average energy of molecular interactions can be decomposed as follows:

$$\mathrm{\Delta }{E}_{\text{bind}}=\mathrm{\Delta }{E}_{\mathrm{v}}+\mathrm{\Delta }{E}_{\mathrm{v}\mathrm{d}\mathrm{w}}+\mathrm{\Delta }{E}_{\text{coul}}+\mathrm{\Delta }{W}_{\mathrm{P}\mathrm{B}}/\mathrm{\Delta }{W}_{\mathrm{G}\mathrm{B}}+\mathrm{\Delta }{W}_{\mathrm{n}\mathrm{p}}$$ (4)

where $\mathrm{\Delta }{E}_{\mathrm{v}}$ represents the changes in valence energy (bond, angle, dihedral, and improper dihedral energies), $\mathrm{\Delta }{E}_{\mathrm{v}\mathrm{d}\mathrm{w}}$ is the change in van der Waals energy, $\mathrm{\Delta }{E}_{\text{coul}}$ is the change in Coulombic energy, $\mathrm{\Delta }{W}_{\mathrm{P}\mathrm{B}}/\mathrm{\Delta }{W}_{\mathrm{G}\mathrm{B}}$ is the electrostatic contribution to the solvation free energy calculated by the Poisson-Boltzmann method or Generalized–Born approximation, respectively, and $\mathrm{\Delta }{W}_{\mathrm{n}\mathrm{p}}$ is the nonpolar contribution to the solvation free energy calculated by the solvent-accessible surface area model.⁴³ The valence energy term in Equation (4) is canceled due to the single trajectory approach. We also report the total electrostatic contribution, $\mathrm{\Delta }{E}_{\mathrm{e}\mathrm{l}}=\mathrm{\Delta }{E}_{\text{coul}}+\mathrm{\Delta }{W}_{\mathrm{P}\mathrm{B}}/\mathrm{\Delta }{W}_{\mathrm{G}\mathrm{B}}$, and the total nonpolar contribution, $\mathrm{\Delta }{E}_{\mathrm{n}\mathrm{p}}=\mathrm{\Delta }{E}_{\mathrm{v}\mathrm{d}\mathrm{w}}+\mathrm{\Delta }{W}_{\mathrm{n}\mathrm{p}}$.

In this work, we consider the change in the binding energy between wild type and mutant forms of the MEIG1 protein forming a protein–protein complex with PACRG, which is the difference between the binding energy of the wild type and mutant computed separately using MMPBSA/MMGBSA:

$$\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}=\mathrm{\Delta }{E}_{\text{wild}\text{type}}-\mathrm{\Delta }{E}_{\text{mutant}}$$ (5)

where $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$ is the change in the binding energy between the wild type MEIG1-PACRG complex binding energy, $\mathrm{\Delta }{E}_{\text{wild}\text{type}}$, and the binding energy of the mutant MEIG1-PACRG complex, $\mathrm{\Delta }{E}_{\text{mutant}}$.

Fpocket⁴⁴ and MDpocket⁴⁵ were used to identify binding pockets on MEIG1 and PACRG proteins. The human and mouse models, including wild type apo-MEIG1, apo-PACRG, and MEIG1–PACRG complex, were used to examine potential ligand binding sites, particularly those involving residues W50 and Y68. Fpocket is a protein pocket prediction algorithm and was used to detect pockets on the final frame of the GaMD simulations. Fpocket input command line arguments were either set to detect small molecule binding sites (default settings), or big external pockets in the case of apo-MEIG1. Results included the druggability score and pocket volume of all binding sites detected. The druggability score is a numerical value between 0 and 1, giving the preference of the pocket to bind a small molecule. The pocket volume is approximated in units of ų. The resulting pockets identified from Fpocket were then analyzed using MDpocket by searching the area of interest for a potential pocket over the last 5 ns of GaMD simulation. MDpocket reported the druggability score and pocket volume of the pocket. MDpocket results were used to calculate the average druggability score, average pocket volume, and the pocket stability; where the pocket stability is the number of structures that show a pocket divided by the total number of structures computed in the MDpocket simulation.

3 |. RESULTS AND DISCUSSION

3.1 |. Mutations of W50A and Y68A in human MEIG1

Both W50A and Y68A mutations have been shown to reduce the complex formation of MEIG1 with PACRG.⁶ In this study, we focus on the protein dynamics around these mutation sites. Our GaMD results indicate that W50 interacts with K57, Y69, and K71 of MEIG1 and H137 of PACRG (Figure 2A), whereas Y68 interacts with K58 and F66 of MEIG1 and W96, I100, and E101 of PACRG (Figure 2B). Figure 2C shows the distance change between W50, Y68, and key residues during GaMD simulations. Overall, after mutations, the distance between two key residues fluctuates more, indicating that both W50A and Y68A mutations destabilize the complex structure. The distance between W50 and K71 decreases during nearly the entire simulation after the bulky tryptophan was mutated to a smaller alanine. Residues I100 and E101 of PACRG show large fluctuations after the Y68A mutation as I100 moves closer to Y68A to fill the space when the Y68 sidechain is replaced with alanine.

Electrostatic interactions play a crucial role in the MEIG1–PACRG system. We explored the hydrogen bonding around W50 and Y68. Our simulations reveal that W50 forms hydrogen bonds with Y69 of MEIG1, and Y68 forms hydrogen bonds with E101 of PACRG (Figure 2A,B). Table 2 shows the change in hydrogen bond occupancy with and without mutations. In the human wild type models, the hydrogen bond occupancy of W50 and Y69 was 57.10% and 56.72% and of Y68 and E101 was 79.43% and 97.84%. After W50A and Y68A mutations, the hydrogen bonds were eliminated. However, W50A and Y68A mutations did not affect the hydrogen bonding of the other nonmutated residue. For example, the W50A mutation does not significantly disrupt the hydrogen bonds between Y68 and E101.

TABLE 2.

Hydrogen bond occupancy of W50–Y69 and Y68–E101 (human)/Y68–E85 (mouse) in the human and mouse MEIG1–PACRG complex.

Model No.	Organism	Mutation	W50–Y69 hydrogen bond occupancy (%)	Y68–E101/E85 hydrogen bond occupancy (%)
1	Human	Wild type	57.10	79.43
2	Human	Wild type	56.72	97.84
3	Human	W50A	0	69.96
4	Human	W50A	0	92.92
5	Human	Y68A	58.38	0
6	Human	Y68A	57.24	0
15	Mouse	Wild type	52.45	78.77
16	Mouse	Wild type	51.31	90.22
17	Mouse	W50A	0	93.69
18	Mouse	W50A	0	95.04
19	Mouse	Y68A	59.74	0
20	Mouse	Y68A	59.91	0

Note: Data is given as the percentage of frames in GaMD trajectories that have hydrogen bonds.

Abbreviations: GaMD, Gaussian-accelerated molecular dynamics; MEIG1, meiosis-expressed gene 1; PACRG, Parkin co-regulated gene.

The MMPBSA and MMGBSA calculations were used to explore the MEIG1–PACRG protein–protein interaction energies over the last 25 ns of simulations. The decomposed energetic terms from the calculations help to identify the energy contribution that leads to the association of MEIG1 with PACRG. Both our MMPBSA (Table 3) and MMGBSA (Table S1) results show that, in the human models, the change in the total interaction energy, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$, between the wild type and two mutants is negative, suggesting that both W50A and Y68A mutations can disturb the MEIG1 and PACRG binding. However, the energetic change of the Y68A mutant is greater than the W50A mutant. In the MMPBSA calculations, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$ is −19.62 and − 19.48 kcal/mol for the Y68A mutation, whereas $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$ is −1.46 and − 2.75 kcal/mol for the W50A system. A similar trend was presented in the MMGBSA calculations, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$ is −18.38 and − 18.76 kcal/mol for Y68A, whereas $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$ is −3.41 and − 4.27 kcal/mol for W50A. Overall, both mutations result in the increase of electrostatic energy $\left(\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{e}\mathrm{l}}\right)$ more than nonpolar energy $\left(\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{n}\mathrm{p}}\right)$, revealing that the polar interaction network around W50/Y68 has significant contributions to the MEIG1–PACRG binding. Moreover, the total electrostatic energy increases more for the Y68A mutation than the W50A mutation, which could be attributed to the hydrogen bond forming between Y68 and E101 and other electrostatic interactions surrounding Y68.

TABLE 3.

Energy (kcal/mol) obtained from MMPBSA calculations.

Model No.	Organism	Mutation	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{vdw}}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{coul}}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{PB}}$	$\mathrm{\Delta }\mathrm{\Delta }{W}_{\mathrm{np}}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{el}}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{np}}$	$\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{bind}}$
1, 3	Human	W50A	−0.40	−21.67	20.54	0.07	−1.13	−0.33	−1.46
2, 4	Human	W50A	−4.82	−37.19	32.33	6.93	−4.86	2.11	−2.75
1, 5	Human	Y68A	−3.95	−28.66	11.40	1.59	−17.26	−2.36	−19.62
2, 6	Human	Y68A	−2.64	−65.20	48.45	−0.09	−16.75	−2.73	−19.48
15, 17	Mouse	W50A	−2.30	−0.93	−3.65	4.80	−4.57	2.50	−2.08
16, 18	Mouse	W50A	0.72	−21.51	14.70	4.58	−6.81	5.30	−1.51
15, 19	Mouse	Y68A	1.77	−86.55	72.45	−4.38	−14.10	−2.62	−16.72
16, 20	Mouse	Y68A	−0.48	−52.27	36.56	−1.72	−15.71	−2.20	−17.91

Note: The energy is computed according to the final 25-ns simulations of human and mouse MEIG1–PACRG complexes. $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{vdw}}$ and $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{coul}}$ are the van der Waals and electrostatic interactions, respectively, between the wild type and mutants. $\mathrm{\Delta }\mathrm{\Delta }{W}_{\mathrm{PB}}$ is the polar solvation free energy calculated by the Poisson–Boltzmann model. $\mathrm{\Delta }\mathrm{\Delta }{W}_{\mathrm{np}}$ is the nonpolar contribution of the solvation energy. $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{el}}$ represents the sum of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{coul}}$ and $\mathrm{\Delta }\mathrm{\Delta }{W}_{\mathrm{PB}}$. $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{np}}$ represents the sum of $\mathrm{\Delta }\mathrm{\Delta }{E}_{\mathrm{vdw}}$ and $\mathrm{\Delta }\mathrm{\Delta }{W}_{\mathrm{np}}$. ΔΔ indicates the changes between the calculations from the mutated and wild type state. The model numbers in the first column indicate the two states used to calculate ΔΔ.

Abbreviations: MEIG1, meiosis-expressed gene 1; MMPBSA, Molecular Mechanics Poisson–Boltzmann Surface Area; PACRG, Parkin co-regulated gene.

Pairwise nonbonded energy was calculated to explore the details of the local interaction energies involving W50 and Y68. We computed pairwise energy for W50 and Y68 separately between their respective nearby interacting residues. Table 4 shows the change in electrostatic and van der Waals energy between wild type and mutants. After both W50A and Y68A mutations, the pairwise energy increases. Moreover, the increase of electrostatic energy is significantly more than van der Waals energy, indicating that the electrostatic interactions around W50 and Y68 play a crucial role in MEIG1–PACRG binding. The polar atoms on the W50 and Y68 sidechain increase the hydrophilicity of the MEIG1–PACRG interface. Losing these electrostatic interactions results in the decrease of complex stability. A similar effect can also be applied to nonpolar sidechains.

TABLE 4.

Pairwise energy (kcal/mol) for human and mouse MEIG1–PACRG complexes.

Model No.	Organism	Mutation	$\mathrm{\Delta }\langle E_{\mathrm{VDW}}\rangle$	$\mathrm{\Delta }\langle E_{\mathrm{Coul}}\rangle$
1, 3	Human	W50A	−6.96	−25.32
2, 4	Human	W50A	−6.50	−50.72
1, 5	Human	Y68A	−10.46	−31.30
2, 6	Human	Y68A	−10.08	−31.47
15, 17	Mouse	W50A	−8.32	−67.30
16, 18	Mouse	W50A	−7.16	−41.54
15, 19	Mouse	Y68A	−10.88	−46.34
16, 20	Mouse	Y68A	−10.45	−30.20

Note: $\mathrm{\Delta }\langle E_{\mathrm{VDW}}\rangle$ and $\mathrm{\Delta }\langle E_{\mathrm{Coul}}\rangle$ are the van der Waals and electrostatic energies, respectively. Models labeled W50A gives the change in pairwise energy between wild type and mutant calculated using residues W50, K57, Y69, K71, and H137. Models labeled Y68A gives the change in pairwise energy between wild type and mutant calculated using residues Y68, K58, F66, W96, I100, and E101 for human and W80, I84, E85, and H121 for mouse.

Abbreviations: MEIG1, meiosis-expressed gene 1; PACRG, Parkin co-regulated gene.

To examine the overall protein motion, we performed RMSF (Figure 3) on GaMD trajectories of the MEIG1 and PACRG complex for both wild type and mutant systems. The RMSF calculations show that the terminal of MEIG1 and PACRG are more flexible than internal protein residues, especially the C-terminal of PACRG. However, the N-terminal of PACRG, residues 79–98 in the loop, is stable and forms direct interactions with MEIG1. Fluctuations were found in MEIG1 residues 42–47 and 51–54 while W50 between the two regions shows a low RMSF less 1. The residues of 42–54 in MEIG1 form a loop connecting α2 and β2, which is usually considered flexible; however, the low RMSF of W50 is attributed to the nonpolar interactions formed by W50, the alkyl sidechain of K57 on the same loop, and K71 on the nearby loop between β3 and α3. A similar effect is also displayed in the MEIG1 residue Y68, that is, the motions of K58, F66, and Y68 are restrained as they form interactions with each other. These same results are found for the wild type and mutant, W50A and Y68A, proteins.

FIGURE 3.

Root mean square fluctuation (RMSF) computed from Parkin co-regulated gene (PACRG) and meiosis-expressed gene 1 (MEIG1) models with the last 25 ns of simulations. Holoprotein wild type, W50A mutant, and Y68A mutant are shown in red, blue, and green, respectively. Apoprotein is shown in purple.

The pairwise correlation between residues of the MEIG1 and PACRG complex was calculated for wild type and mutant systems using correlation analysis. The produced correlation maps show strong correlations between residues 50 and 71 in MEIG1 (Figure 4 box), while weaker correlations are presented between PACRG residues. W50 is highly correlated with K57, Y69, and K71 with correlations >0.65; and Y68 is highly correlated with K58 and F66 with correlations >0.70 (Table S2), indicating the synchronous motion of these interacting residues. Correlations of W50-H137, Y68-W96, Y68-I100, and Y68-E101 are positive and vary from 0.36 to 0.47. In the human MEIG-PACRG complex, both RMSF and correlation of MEIG1 do not change significantly after the W50A and Y68A mutations, suggesting that the global protein dynamics are similar before and after the mutations even though the binding energies, local pairwise energies, atom–atom distance, and hydrogen bonding alter substantially.

FIGURE 4.

Correlation maps computed from human meiosis-expressed gene 1 (MEIG1)–Parkin co-regulated geneParkin co-regulated gene (PACRG) complex simulations. Correlation of MEIG1 is shown on the top with holo-MEIG1 wild type and mutants, and apo-MEIG1 wild type. PACRG is shown on the bottom with holo-PACRG from wild type and mutant simulations, and apo-PACRG. The black box in Model 2 highlights the correlations between residues 50 and 71 of MEIG1. The maximum value of 1.0 represents highly correlated motions of two residues, whereas the minimum value of −1.0 indicates inverse correlation.

3.2 |. Mutations of W50A and Y68A in mouse MEIG1

The sequence identity between mouse and human MEIG1 is 87.06%, and between mouse and human PACRG is 81.68%. All key residues interacting with W50 and Y68 in the mouse models are the same in the human one. For example, in the mouse model, W50 interacts with K57, Y69, and K71 of MEIG1 and H121 of PACRG (Figure S1A) while Y68 interacts with K58 and F66 of MEIG1 and W80, I84, and E85 of PACRG (Figure S1B). Because mouse and human models show high sequence identity and similar interactions around the protein–protein interface, we expect that the mouse and human protein may demonstrate similar protein dynamics.

The distance between W50/Y68 and key residues was computed in the mouse model. In the wild type MEIG1 and PACRG, these distances fluctuate less than the mutants (Figure S1C), indicating that the W50A and Y68A mutations destabilize the mouse complex, similar to the effect shown in the human model. Furthermore, the distance between each nonbonded pair is nearly identical for the human and the mouse models, except for H121. For example, the average distance between W50 and K57 is 5.07 Å and 5.09 Å in the human and mouse model, respectively. In the mouse model, the distance fluctuation between wild type W50 and H121 is about 7 Å, and between mutant W50A and H121 is about 12 Å, whereas the fluctuation in wild type and mutant human models is <1 Å throughout the simulation. The enhanced fluctuation from the mouse model can be attributed to the rotation of the H121 sidechain between two conformations, one with the imidazole ring pointing toward W50 (about 8 Å) and the other pointing away (about 12 Å; Figure S2). However, near the end of simulation, in the last 25 ns, this distance settles down to about 8 Å same as the distance for human models.

Hydrogen bond analysis was performed on the mouse models to explore the electrostatic interactions around W50 and Y68. Similar to the human model, W50 and Y68 in the mouse MEIG1 form hydrogen bonds with Y69 and E85, respectively (Figure S1). Also, all hydrogen bonds were eliminated after W50A and Y68A mutations. However, overall, the hydrogen bond occupancy of W50 and Y68 in mouse models is lower than the human, for example, 52.45% and 51.31% occupancy was found in mouse W50–Y69 in contrast with 57.10% and 56.72% in human (Table 2), displaying that the mouse models have slightly weaker hydrogen bonding for W50 and Y68 although the hydrogen bonding network of mouse and human models is the same.

MMPBSA (Table 3) and MMGBSA (Table S1) calculations were performed to study the energetic changes of the W50A and Y68A mutation in the mouse models. Generally, the energetic changes of the mouse model are similar to what has been reported in the human model—(1) both W50 and Y68 mutations abolish the MEIG1–PACRG binding as a result of unfavorable energies, (2) Y68 contributes more than W50 in the protein–protein interaction because the total binding energy, $\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{bind}}$, of the Y68A mutation increases more, and (3) electrostatic energy plays a greater role in the Y68A mutation than nonpolar energy as a strong polar interaction network is associated with Y68 and E85. The computed energies agree with the experimental findings that, although both W50A and Y68A mutations in mouse MEIG1 disrupt MEIG1–PACRG binding, the Y68A mutation interrupts the complex formation more effectively than W50A.⁷ Note that part of the mouse model (PACRG) was constructed by homology modeling. MMPBSA/GBSA performed using the last 25 ns of simulation still reproduces the energetic results computed from the human model, showing that the mouse complex structure is well-equilibrated after 175 ns of GaMD simulation, in which the trajectories can be further used for drug discovery, such as compound screening.

Pairwise energy calculation was used to reveal interaction energies of the key nonbonded interacting partners around W50 and Y68 in the mouse. As expected, the energies increase after W50A and Y68A mutations, expressing unfavorable conformations in mutants (Table 4). The calculations also show that electrostatic interactions contribute more than nonpolar interactions in the MEIG1–PACRG binding as both W50A and Y68A mutations result in the energy increase of electrostatic interactions more than van der Waals. These results are consistent with the energy calculations using human models.

To compare the protein flexibility and residue fluctuation between human and mouse models, RMSF was calculated using the mouse trajectories. Generally, the mouse and human complex display similar RMSF values and protein dynamics (Figure 3)—with small fluctuations found for residues involved in MEIG1–PACRG binding (e.g., W50 and Y68) and larger fluctuations away from the interaction site. The RMSF similarity between mouse and human holoproteins is seen for wild type and mutant systems, but only once the mouse system is well equilibrated. The mouse protein demonstrates larger fluctuations during the first 175-ns simulations than the last 25 ns (Figure S3). When compared with MEIG1, the mouse PACRG particularly displays larger atomic fluctuations at the beginning of simulations (Figure S4). This is a result of the initial mouse PACRG structure predicted by homology modeling requiring longer equilibration than the human structure, which is consistent with the results reported from MMPBSA calculations.

Correlation analysis (Figure S5) was performed on the mouse models to explore the motions between protein residue pairs. The correlation maps computed from the mouse are similar to human models for both holo-MEIG1 and holo-PACRG. In mouse models, residues 50–71 of MEIG1 show strong correlations (Figure S5, box). W50 is strongly correlated with K57, Y69, and K71, and Y68 is strongly correlated with K58 and F66. The correlation values of these residues are >0.70 (Table S3). Other residues interacting with W50 and Y68 also show moderate correlations, for example, the correlation of the residue pairs of W50–H121, Y68–W80, Y68–I84, and Y68–E85 ranges from 0.27 to 0.43 (Table S3). Similar to human mutants, the W50A and Y68A mutations do not alter the correlation significantly. This correlation similarity between mouse and human models indicates that the complex structures of the mouse and human have similar protein dynamics involving W50 and Y68. Although the mouse MEIG1–PACRG complex has less stable H121 near the binding site and slightly weaker hydrogen bonding compared with human, the mouse and human models are analogous in most aspects.

3.3 |. Dynamics of apoprotein and holoprotein

RMSF was calculated on equilibrated protein systems to reveal the change in protein dynamics before and after complex formation. Overall, the RMSF and protein dynamics of apoprotein and holoprotein are similar, although slight variations are found near the N-terminal and C-terminal areas (Figure 3). Theoretically, proteins become rigid after binding; however, the terminal regions of MEIG1 are more flexible when the complex forms. This can be attributed to the restricted motions of the residues on the protein–protein interface after binding, so that the interactions between the internal residues at the interface remain stable while fluctuations of the terminal regions are enhanced. In addition, compared with apo-MEIG1, as expected, apo-PACRG takes a longer time to reach equilibration, especially in the region of residues 190–210, which is consistent with the findings from holo-PACRG (Figures S3 and S4).

RMSF of individual residues was examined to determine the effect of the protein binding around W50 and Y68. Our RMSF calculations show that, in both human and mouse models, the residues around W50 fluctuate more after binding, while the flexibility of the residues surrounding Y68 reduces (Figure S6). In the apo-MEIG1, nonpolar interactions with W50, K57, and K71 stabilize the local motion of W50. After binding, W50, K57, and K71 can alternatively interact with residues from PACRG (e.g., H137) which results in the RMSF increase of these three residues. However, Y68 demonstrates a different binding mode. After the complex forms, Y68 interacts with K58 and F66 of MEIG1 and W96, I100, and E101 of PACRG. The strong polar interactions across the MEIG1–PACRG interface near Y68 stabilize the binding. In the apo forms, these residues do not interact strongly with each other; instead, they fluctuate with water a molecule, which explains the RMSF decrease after binding. RMSD was also calculated on the residues 50–71 of MEIG1 and residues 94–101 of PACRG to further investigate the motions at the MEIG1–PACRG interaction interface (Figure S7). In both human and mouse systems, RMSD of interfacial residues reduces, indicating that the MEIG1–PACRG binding stabilizes the local interfacial interactions involving W50, Y68, and nearby key residues.

We constructed correlation maps to explore the correlated motion changes between apoprotein and holoprotein. In holo-MEIG1, correlation is found between interfacial residues 50–71; however, this does not show in the apo-MEIG1 (Figure 4). In both human and mouse models, the correlations of the resides around W50 and Y68 are enhanced, ranging from 0.15 to 0.5, after the complex formation. For example, the correlation between W50 and K57 in human apo-MEIG1 and holo-MEIG1 is 0.46 and 0.78, respectively (Figures S2 and S4). This suggests that the correlated motions of MEIG1 residues 50–71 are created by the binding of PACRG. Although holo-PACRG did not show strong correlations between each residue pair, the correlations in apo-PACRG decrease, particularly in the regions composed of W96, I100, E101, and H137. This correlation difference is found in both human and mouse systems.

3.4 |. Ligand binding site prediction

Binding pocket identification and analysis were performed using Fpocket and MDpocket to discover potential ligand binding sites near W50 and Y68. In the human and mouse complex, 10 and 9 potential pockets have been identified, respectively (Figures 5A and S8A). Because impeding the complex formation is a strategy to develop male-based contraception, we particularly focus on the pockets along the MEIG1–PACRG interface. Although many pockets at various locations on the protein surface or within small cavities were found, we could locate either one or two pockets on the protein–protein binding surface from crystal structures and MD snapshots (Tables S6 and S7). The pockets close to W50 and Y68 generally show high pocket rank or good drug score, and the average volume of these pockets ranges from 90 to 500 ų (Table S7), indicating the potential to bind a drug-like molecule.

FIGURE 5.

Fpocket results of human meiosis-expressed gene 1 (magenta) in complex with Parkin co-regulated gene (green). (A) The primary binding pocket (orange) and all potential binding pockets (gray) predicted by Fpocket. (B) The zoomed image of the primary binding pocket at the interface involving Y68.

In the human complex, we found a pocket that involves key residues from both MEIG1 (K57, K58, and Y68) and PACRG (I100, E101, and H137; Figure 5B). This pocket can be identified in the MD trajectories with an average score of more than 0.26 (Table S7). In the human apo-MEIG1 and apo-PACRG proteins, the pockets are displayed near those key residues as well, suggesting that the pocket shown in Figure 5B could be a target site in drug development. In the mouse model, we found a pocket similar to the human protein, which involves W50, K57, K58, Y68, and Y69 of MEIG1 and I84, E85, and H122 of PACRG (Figure S8B). This pocket shows a good average score (>0.28) and high stability (>0.96) in the MD simulations (Table S7). A similar pocket can also be detected in the apo mouse proteins.

When compared with apo models, more pockets on the protein binding surface were identified from holoproteins. Also, the average scores of the pockets from the complexes are better than the apoproteins (Table S7). As expected, residues on the protein surface, such as K58 and Y68, are more flexible before binding, which results in difficulty in pocket prediction in the apoproteins. In addition, potential binding pockets of both complexes and apo-PACRG proteins can be detected using a standard search method; however, some pockets near W50/Y68 in apo-MEIG1 only show when the large, external pocket search was applied. The average score of the pockets from apo-MEIG1 is lower than the one from apo-PACRG. This could be attributed to the primary binding residues W50 and Y68 from apo-MEIG1 on the loops between α2 and β2, and between β3 and α3, respectively. The flexibility of the loops makes the pockets in apo-MEIG1 less stable. However, the major protein–protein binding regions of PACRG, α2, α3, and the loop between α1 and α2 (Figure 1), are stable. Hence, if a stable pocket is of interest in drug design, PACRG is a better target than MEIG1.

4 |. CONCLUSIONS

In this study, we investigated the dynamic properties of the MEIG1 and PACRG protein that are essential for the development of mature sperm cells through the spermiogenesis process. GaMD simulations were carried out on apo-MEIG1, apo-PACRG, and the MEIG1–PACRG complex. Both human and mouse models were considered in this work. For all systems, the wild-type protein and two mutants, W50A and Y68A, were studied in order to reveal the critical role of W50 and Y68 in MEIG1–PACRG binding. Our results show that the interactions between W50/Y68 and nearby residues on the MEIG1–PACRG interface become less stable after the W50A and Y68A mutations. The energy calculations also show that, compared with the W50A mutant, Y68A results in a significantly unfavorable energetic penalty in the protein–protein association because polar interactions between Y68 and PACRG interfacial residues, including Y68–E101/E85 hydrogen bonds, reduce. This agrees well with the experimental finding.⁷ Moreover, human and mouse models demonstrate similar protein dynamics. Although the mouse models took a long time in equilibration as the mouse PACRG protein is built by homology modeling, after more than a 100 ns of GaMD simulation, the mouse models display similar atomic fluctuations and energetic properties to the human models. Therefore, the mouse system of MEIG1, PACRG, and the complex could be a good model used for in vivo experimental testing of drug binding and ligand design. By studying the apoprotein and holoprotein, we found that the association of MEIG1 with PACRG results in the change of correlated motions on interfacial residues. Upon binding, the sidechains of the interfacial residues adjust to stabilize the local interaction network between MEIG1 and PACRG, whereas the backbones of the residues do not change substantially. To guide future male-based contraception development, we reported 10 potential ligand binding pockets according to our MD trajectories. We found a pocket involving W50 and Y68 that is of particular interest in drug design to inhibit the MEIG1–PACRG interaction. Investigation of this pocket through virtual screening and molecular docking could lead to identifying ligands that can potentially inhibit the association of MEIG1 and PACRG, with the end goal of drug discovery for a male-based contraceptive.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.