Inhibitor binding and disruption of coupled motions in MmpL3 protein: Unraveling the mechanism of trehalose monomycolate transport

Abstract

Mycobacterial membrane protein Large 3 (MmpL3) of Mycobacterium tuberculosis (Mtb) is crucial for the translocation of trehalose monomycolate (TMM) across the inner bacterial cell membrane, making it a promising target for anti‐tuberculosis (TB) drug development. While several structural, microbiological, and in vitro studies have provided significant insights, the precise mechanisms underlying TMM transport by MmpL3 and its inhibition remain incompletely understood at the atomic level. In this study, molecular dynamic (MD) simulations for the apo form and seven inhibitor‐bound forms of Mtb MmpL3 were carried out to obtain a thorough comprehension of the protein's dynamics and function. MD simulations revealed that the seven inhibitors in this work stably bind to the central channel of the transmembrane domain and primarily forming hydrogen bonds with ASP251, ASP640, or both residues. Through dynamical cross‐correlation matrix and principal component analysis analyses, several types of coupled motions between different domains were observed in the apo state, and distinct conformational states were identified using Markov state model analysis. These coupled motions and varied conformational states likely contribute to the transport of TMM. However, simulations of inhibitor‐bound MmpL3 showed an enlargement of the proton channel, potentially disrupting coupled motions. This indicates that inhibitors may impair MmpL3's transport function by directly blocking the proton channel, thereby hindering coordinated domain movements and indirectly affecting TMM translocation.

1. INTRODUCTION

Tuberculosis (TB), an infectious disease caused by Mycobacterium tuberculosis (Mtb), remains a major global health threat in the 21st century. According to the World Health Organization (WHO) TB report (WHO, ²⁰²³), in 2022, an estimated 7.5 million people were newly diagnosed with TB globally, and around 1.3 million died from the disease, making it one of the leading causes of death worldwide. The increasing prevalence of multidrug‐resistant TB and extensively drug‐resistant TB poses a significant obstacle to global TB control efforts (Ormerod, 2005; Raviglione & Smith, ²⁰⁰⁷; WHO, ²⁰²³). Therefore, there is an urgent need for the development of novel therapeutic strategies targeting new mechanisms of action to overcome drug resistance and improve treatment outcomes.

The unique and complex multilayered membrane architecture of the Mtb cell wall plays a crucial role in the pathogenesis of TB (Angala et al., 2014; Karakousis et al., ²⁰⁰⁴). This structure not only provides Mtb with a formidable barrier against host immune defenses but also confers resistance to clinically relevant antibiotics, essential for its survival. The outer membrane (OM) is primarily composed of long‐chain fatty acids, known as mycolic acids (MAs), ranging from C60 to C90. These MAs traverse the inner membrane (IM) as trehalose monomycolates (TMMs) and then undergo one of two pathways: either covalently linking to the arabinogalactan‐peptidoglycan layer to form mycoloyl‐arabinogalactan‐peptidoglycan or converting into trehalose dimycolates (TDM) (Brennan, 2003; Brennan & Nikaido, ¹⁹⁹⁵; Marrakchi et al., ²⁰¹⁴; Takayama et al., ²⁰⁰⁵).

Mycobacterial membrane protein Large 3 (MmpL3), a member of the resistance‐nodulation‐division (RND) superfamily, a proton motive force (PMF)‐dependent transporter protein, is tasked with the translocation of TMMs from the cytoplasmic side to the periplasmic side in Mtb (Grzegorzewicz et al., 2012; Su et al., ²⁰¹⁹; Su et al., ²⁰²¹; Varela et al., ²⁰¹²; Xu et al., ²⁰¹⁷). Inhibiting the transport pathway mediated by MmpL3 effectively disrupts the biosynthesis of TDM, consequently compromising the integrity of the bacterial OM (Grzegorzewicz et al., 2012; Varela et al., ²⁰¹²). Structural biology studies show that MmpL3 protein is composed of three primary structural domains: a periplasmic pore domain (PD), further divided into the PN and PC subdomains; an unstable long C‐terminal domain; and a 12‐helix transmembrane domain (TMD) responsible for proton transportation (Adams et al., 2021; Couston et al., ²⁰²³; Hu et al., ²⁰²²; Su et al., ²⁰¹⁹; Su et al., ²⁰²¹; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹) (Figure 1a). Within the TMD, two critical “ASP‐TYR” pairs (ASP251‐TYR641 and ASP640‐TYR252 in Mtb, ASP256‐TYR646 and ASP645‐TYR257 in M. smegmatis) located on the fourth and tenth helices play a vital role in the intracellular proton transfer pathway, highlighting its central function in cellular physiology (Bernut et al., 2016; Grzegorzewicz et al., ²⁰¹²; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹). The spheroplast assay has suggested that MmpL3 functions as a TMM flippase, facilitating the translocation of TMM from the inner to the outer leaflet of the IM (Xu et al., 2017). Furthermore, the M. smegmatis MmpL3 structure has unveiled two TMM binding sites, providing significant insights for understanding the TMM transport mechanism from the outer leaflet of the IM to the PD (Su et al., 2021).

FIGURE 1.

MD simulation systems. (a) Structure of Mtb MmpL3‐APO embedded in the mycobacterial IM, MmpL3 is modeled through homology modeling. Two critical “ASP‐TYR” pairs (ASP251‐TYR641 and ASP640‐TYR252) in the TMD are highlighted. (b) Structure of Mtb MmpL3‐SQ109 embedded in the mycobacterial IM, the binding pocket of SQ109 is highlighted. (c) Chemical structures of inhibitors studied in this research. MmpL3 is shown in a cartoon representation while lipids are shown by sticks.

Given its indispensable role in bacterial survival, MmpL3 is considered an attractive target for the development of novel anti‐TB drugs (Imran et al., 2022; North et al., ²⁰²³; Shao et al., ²⁰²⁰). The feasibility of this strategy has been preliminarily validated through the discovery and investigation of a series of MmpL3 inhibitors, including AU1235 (Grzegorzewicz et al., 2012), SPIRO (Ray et al., 2021), NITD349 (Rao et al., 2013), Rimonabant (Ramesh et al., 2016), and the clinically advanced SQ109 (Imran et al., 2022; Tahlan et al., ²⁰¹²), currently in phase 2b/3 trials (Figure 1b,c). In addition, some MmpL3 inhibitors have demonstrated potential synergistic effects with existing anti‐TB drugs (Reddy Venkata et al., 2010; Stec et al., ²⁰¹⁶). Multiple high‐resolution structures of M. smegmatis MmpL3 in complex with inhibitors have provided an opportunity for structure‐based drug design, exemplified by compounds such as ST004 (Hu et al., 2022). Detailed structural analysis of M. smegmatis has shown that inhibitor binding disrupts the hydrogen bond network of the “ASP‐TYR” pairs at the central channel of the TMD, blocking the proton transport process (Adams et al., 2021; Hu et al., ²⁰²²; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹) (Figure 1b). However, the majority of identified MmpL3 inhibitors exhibit pronounced hydrophobic characteristics, which may be in response to hydrophobic pockets present in MmpL3, potentially impeding their further development to some extent (Hu et al., 2022; North et al., ²⁰²³; Poce et al., ²⁰¹⁶; Shao et al., ²⁰²⁰; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹). While various studies have provided significant insights into the transport mechanism of TMMs and the inhibition mechanism of MmpL3 by inhibitors (Carbone et al., 2023; Stevens et al., ²⁰²²; Su et al., ²⁰¹⁹; Su et al., ²⁰²¹; Xu et al., ²⁰¹⁷; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹), the mechanisms at the atomic level are still not fully understood. Hence, a further thorough comprehension of the transport mechanism of TMM and the inhibitory mechanism of MmpL3 inhibitors can provide a solid scientific foundation for developing novel and effective anti‐TB drugs targeting Mtb MmpL3.

In this study, in order to uncover the mechanisms underlying TMM transport and inhibition of Mtb MmpL3, molecular dynamic (MD) simulations for apo form and seven inhibitor‐bound forms of Mtb MmpL3 were carried out. The initial structure of Mtb MmpL3 was constructed through homology modeling using M. smegmatis MmpL3. MD simulations were performed to reveal their dynamic behavior and complex interactions of protein and inhibitors at the atomic level, thus offering deep insights into protein functionality and the mechanism of action of inhibitors. Based on the obtained trajectories, the root‐mean‐square deviation (RMSD) and root‐mean‐square fluctuation (RMSF) were used to characterize conformation changes of MmpL3 and identify flexible regions within the protein, respectively. The molecular mechanics‐generalized Born surface area (MM/GBSA) method and hydrogen bonds analysis were applied to characterize the interactions between the inhibitors and MmpL3. With the structural and energy analysis, the binding modes of the inhibitors and the key residues involved in inhibitor binding were identified. Furthermore, the dynamical cross‐correlation matrix (DCCM) was used to determine the potential dynamics correlations. principal component analysis (PCA) was used to discover the principal motion modes. In addition, distance calculations were performed to accurately depict the conformation changes of MmpL3, and Markov state model (MSM) analysis was performed to probe the kinetics of apo form of MmpL3. Based on the integrated analysis above, we have proposed potential mechanisms for TMM transport by MmpL3 and its inhibition.

2. RESULTS AND DISCUSSION

2.1. Homology modeling of Mtb MmpL3

In this study, the structures of Mtb MmpL3 were derived through homology modeling, utilizing the structures of M. smegmatis MmpL3 as templates. The Mtb MmpL3 protein, available in the RCSB Protein Data Bank (PDB), is exclusively represented by the unbound apo form cryo‐electron microscopy structure (PDB ID: 7NVH) (Adams et al., 2021), with residues 343–377 between TMD1 and TMD2 missing. MmpL3 proteins derived from both Mtb and M. smegmatis demonstrate substantial sequence homology, with a notable 72.6% identity observed across 730 residues. Additionally, studies suggest functional interchangeability between Mtb MmpL3 and M. smegmatis MmpL3 (Grzegorzewicz et al., 2012). The PDB database includes several three‐dimensional structures of M. smegmatis MmpL3 bound to inhibitors (Table S1). Furthermore, the X‐ray crystallographic structure of the apo form of M. smegmatis MmpL3 exhibits an intact loop region between TMD1 and TMD2 (PDB ID: 7K7M) (Su et al., 2021). Therefore, to ensure the initial complex structure closely approximates the experimental structure, both the apo form and inhibitor‐bound forms of M. smegmatis MmpL3 were used as templates for homology modeling to construct the three‐dimensional structure of Mtb MmpL3 (Section 4.1). The positional error estimates and the Ramachandran plots for the homology‐modeled structure of Mtb MmpL3 are presented in Figures S1 and S2, respectively. The apo form of Mtb MmpL3 obtained through homology modeling shows significant similarity to the experimentally resolved structures of Mtb MmpL3, with Cα RMSD values of 0.755 Å (Figure S3). Finally, a total of eight systems were constructed, comprising one apo form and seven inhibitor‐bound forms of Mtb MmpL3 (Table S2), and these systems will be subjected to long‐term conventional MD simulations.

2.2. Convergence and reproducibility for MD simulations systems

The RMSD for MmpL3 Cα atoms and the heavy atoms of the inhibitors, relative to the initial structure, was calculated to monitor the equilibrium of each system using CPPTRAJ (Roe & Cheatham III, 2013). In all simulated systems, the overall protein structure (residues 1–730) rapidly reaches equilibrium, with RMSD values of approximately 3 Å, as depicted in Figure 2. Figures S4 and S5 show the RMSD of the PD (residues 35–164, 432–542) and the TMD (residues 10–34, 165–337, 400–415, 543–725), respectively. The RMSD of the TMD quickly stabilizes, while the RMSD of the PD converges after 50 ns. Remarkably, the RMSD of the PD, approximately 2–4 Å, is larger than that of the TMD, which fluctuates around 2 Å. This indicates that the PD is more flexible than the TMD, with or without inhibitor binding. From Figure S6, the two trajectories of ICA38 exhibit significant fluctuations, with maximum RMSD values exceeding 5 Å, eventually stabilizing in the last 50 ns. Meanwhile, other inhibitors rapidly reach equilibrium, with smaller RMSD values of approximately 3 Å, except for NITD349, which exhibits slightly larger RMSD values, around 5 Å. However, the inhibitor‐binding pocket shows relatively modest conformation changes, with RMSD fluctuating around 1 Å as shown in Figure S7. This suggests that the binding of inhibitors does not significantly affect the conformation of the binding pocket.

FIGURE 2.

RMSD of Cα atoms for full‐length MmpL3 from MD simulations, showcasing different systems: (a) MmpL3‐APO, (b) MmpL3‐AU1235, (c) MmpL3‐ICA38, (d) MmpL3‐NITD349, (e) MmpL3‐Rimonabant, (f) MmpL3‐SPIRO, (g) MmpL3‐SQ109, and (h) MmpL3‐ST004. Different colors represent parallel trajectories.

In addition to investigating global and local conformational flexibility, we conducted RMSF analysis of Cα atoms to depict the fluctuations of individual residues throughout the simulation duration. Figure S8 shows the RMSF values display consistent trends across all systems. It can be seen that the PD domain (PN and PC) is highly flexible, whereas the TMD domain (TMD1 and TMD2) remains stable, which is consistent with the RMSD fluctuations (Figures S4 and S5). Additionally, the DSSP (Dictionary of Protein Secondary Structure) analysis was performed, revealing no significant overall shift in the secondary structure across all systems (Figure S9).

2.3. Conserved binding modes among diverse inhibitors and crucial residues in inhibitor binding

Clustering analysis was performed using CPPTRAJ with the k‐means algorithm to identify conformation clusters associated with different inhibitors. This analysis categorized the conformations obtained from MD simulations into 10 distinct groups. The most representative structures are illustrated in Figure S10. All inhibitors bind within the central channel of the TMD, surrounded by transmembrane helices TM4, TM5, TM6, TM10, TM11, and TM12. As shown in Figure S10, the tri‐fluorophenyl group of AU1235, the indole group of ICA38 and NITD349, the benzodioxane of SPIRO, and the geranyl tail of SQ109 accommodate the upper part of the pocket, interacting with hydrophobic residues primarily composed of ILE248, LEU633, and LEU637. The two nitrogen atoms in the urea moiety of AU1235, the amide nitrogen and the indole nitrogen of ICA38 and NITD349, the nitrogen atom in the piperidine of SPIRO, and the amino groups of SQ109 form hydrogen bonds with ASP251 or ASP640. The adamantane moiety of AU1235, the carbocyclic spiro group of ICA38, the dimethylcyclohexane group of NITD349, the spirocyclic thienopyran of SPIRO, the adamantane group of SQ109 occupy the lower part of the pocket, interacting with hydrophobic residues, predominantly involving PHE255 and PHE644. Among these, ICA38 and NITD349 exhibit distinct binding conformations within the binding pocket. For ICA38, a significant positional shift of the indole group occurs, transitioning from forming hydrogen bonds with ASP251 to ASP640, as depicted in Figure S10B. For NITD349, the indole group undergoes a near 180° rotation, altering its hydrogen bonding from ASP640 to ASP251, as illustrated in Figure S10C. Since the Cα RMSD of the pocket does not change significantly (Figure S7), the variations in residue side chains may assist in the conformation changes of the inhibitors. Furthermore, structural superimposition of the experimental structures of the M. smegmatis MmpL3‐NITD349 (PDB ID: 7C2M) and MmpL3‐ICA38 (PDB ID: 6AJJ) complexes reveals that, while their chemical structures are similar, the orientation of the indole groups within MmpL3 differs (Figure S11). While Rimonabant binds within the central channel of the TMD, it occupies slightly different subpockets in the upper part of the pocket, interacting with LEU243, ILE244, VAL634, VAL681, and LEU703. Moreover, the piperidine ring occupies the “ASP‐TYR” pair positions. Rimonabant does not form any hydrogen bonds with MmpL3. ST004, which combines scaffolds from SQ109 and Rimonabant, exhibits partially similar binding modes to the two inhibitors. The binding modes of Mtb MmpL3 with inhibitors reported in this study closely resemble the experimentally resolved structures of MmpL3 from M. smegmatis (Hu et al., 2022; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹).

To assess the binding affinity of different inhibitors to Mtb MmpL3, the MM/GBSA method was utilized to estimate the binding free energy. For each complex, 1000 frames were extracted from the last 50 ns trajectories. To calculate entropies from truncated systems (Genheden et al., 2012), a truncation cutoff of 8 Å beyond the ligand boundary was established, with the dielectric constant set to 1. We selected 20 frames from the last 50 ns of the trajectories for the entropy calculation. Table 1 indicates that the binding free energies (ΔG _pre ) estimated by including the truncated‐NMA entropies (TΔS) improve the performance for MM/GBSA (ΔH _pre ). However, the calculated binding free energy values show significant discrepancies when compared to experimental results. SPIRO, SQ109, and ST004 exhibit higher binding free energies, measured at −31.05 ± 6.42 kcal/mol, −59.90 ± 13.93 kcal/mol, and − 41.92 ± 8.28 kcal/mol, respectively, compared to AU1235 (−24.30 ± 6.97 kcal/mol), ICA38 (−17.85 ± 6.23 kcal/mol), NITD349 (−16.34 ± 7.67 kcal/mol). The MM/GBSA method is typically effective in providing insights for the ranking of substituents within a consistent ligand series (Weng et al., 2019; Zhang et al., ²⁰²²). In our systems, ICA38 and NITD349 exhibit structural similarities, leading to comparable binding free energy values. However, for other systems with significant structural variations, directly ranking binding free energies to evaluate binding affinity might be challenging.

TABLE 1.

Calculated binding free energy and detailed contribution of different energy terms (kcal/mol).

Contribution	AU1235	ICA38	NITD349	Rimonabant	SPIRO	SQ109	ST004
Kd a (μM)	0.29 ± 0.04	0.16 ± 0.02	0.05	41.4 ± 5.2	0.8	1.65 ± 0.2	9.58
ΔE vdw	−45.89 ± 2.84	−51.24 ± 2.49	−45.57 ± 4.36	−58.5 ± 2.91	−60.14 ± 2.84	−73.3 ± 4.88	−80.85 ± 4.83
ΔE ele	−20.52 ± 2.88	−12.64 ± 4.82	−23.75 ± 5.02	−1.37 ± 2.42	−12.72 ± 10.08	−95.65 ± 23.86	−18.79 ± 10
ΔG GB	31.84 ± 2.28	34.6 ± 4.66	40.8 ± 5.04	23.26 ± 2.53	27.64 ± 10.47	89.63 ± 20.34	40.93 ± 7.8
ΔG SA	−5.58 ± 0.16	−6.29 ± 0.18	−5.74 ± 0.26	−7.24 ± 0.27	−6.28 ± 0.21	−7.37 ± 0.25	−8.99 ± 0.28
ΔE MM	−66.41 ± 3.21	−63.88 ± 4.59	−69.31 ± 5.31	−59.87 ± 3.41	−72.87 ± 10.25	−168.95 ± 24.73	−99.63 ± 9.45
ΔG sol	26.26 ± 2.29	28.31 ± 4.65	35.06 ± 5.04	16.01 ± 2.52	21.36 ± 10.44	82.27 ± 20.4	31.94 ± 7.9
ΔH pre	−40.15 ± 3.42	−35.56 ± 3.57	−34.25 ± 5.17	−43.85 ± 5.43	−51.51 ± 4.12	−86.68 ± 12.40	−67.70 ± 6.08
TΔS	−15.85 ± 6.07	−17.71 ± 5.11	−17.91 ± 5.66	−20.21 ± 5.34	−20.46 ± 4.92	−26.78 ± 6.340	−25.78 ± 5.62
ΔG pre	−24.30 ± 6.97	−17.85 ± 6.23	−16.34 ± 7.67	−23.64 ± 7.62	−31.05 ± 6.42	−59.90 ± 13.93	−41.92 ± 8.28
ΔG exp b	−8.91	−9.27	−9.96	−5.98	−8.31	−7.88	−6.84

^a K_d represents the dissociation constant.

^b ∆G _exp  = −RT ln K_d, where R = 8.314 J/(mol*K), T = 298.15 K.

By evaluating the contributions of individual energy components, we found that the gas phase energy (ΔE _MM ) by Van der Waals interactions (ΔE _vdw ) and electrostatic interactions (ΔE _ele ) is the driving force for the binding of all inhibitors compared to solvation‐free energy (ΔG _sol ). Among these, ΔE _vdw contributes more significantly except for SQ109. The protonated inhibitors (SPIRO, SQ109, and ST004) demonstrate significantly larger ΔE _MM values than other non‐protonated inhibitors. All inhibitors show small and similar non‐polar solvation‐free energy (ΔG _SA ) values, indicating that ΔG _sol is more influenced by the polar component of the solvation energy ΔG _GB . For SQ109, the polar interaction terms, ΔE _ele and ΔG _GB , are significantly greater than the other molecules, indicating the substantial impact of the diprotonated ethylenediamine unit on the binding affinity.

Through a detailed binding free energy decomposition analysis, several key residues that contribute more than 1 kcal/mol to the binding energy are identified, as shown in Figure 3 and listed in Table S3. Residues ILE248, ASP251, TYR252, PHE255, SER288, ILE292, LEU633, LEU637, ASP640, TYR641, and PHE644 display significant energy contributions when interacting with more than three inhibitors. Additionally, LEU243, ILE244, GLY247, VAL285, VAL634, ILE674, VAL681, and LEU703 exhibit strong binding free energy contributions to less inhibitors. Most of these residues are hydrophobic, with some contributions nearing 3 kcal/mol, such as for ILE248, PHE255, LEU637, and PHE644. This suggests that hydrophobic interactions play a critical role in the binding of inhibitors. Notably, ASP251, TYR252, ASP640, and TYR641 are situated in the TMD central proton channel, important for proton transfer processes (Hu et al., 2022; Yang et al., ²⁰²⁰; Zhang, Li, et al., ²⁰¹⁹). Hydrogen bond analysis (Table S4) shows that inhibitors primarily form hydrogen bonds with ASP251, ASP640, or both residues. In detail, AU1235 and SPIRO form hydrogen bonds with ASP640; ICA38 and NITD349 with ASP251, ASP640, and TYR641; SQ109 with ASP251 and ASP640; and ST004 primarily with ASP251. In the work of Dupont et al., docking studies revealed that PIPD1 binds to the same cavity but forms slight different hydrogen bonds compared to the inhibitors in this study, specifically with SER288 and TYR641 (Dupont et al., 2019). Furthermore, mutations in ILE244 (Williams John et al., 2019), PHE255 (McNeil Matthew et al., 2020), VAL285 (Korycka‐Machała et al., 2019), SER288 (Li et al., 2020), ILE292 (Dupont et al., 2019), PHE644 (McNeil Matthew et al., 2020), and VAL681 (Dupont et al., 2019) may facilitate the emergence of drug resistance in Mtb, emphasizing the critical role of these residues in inhibitor binding.

FIGURE 3.

Total energy contributions of each residue to the binding free energy for diverse inhibitors, calculated by MM/GBSA: (a) MmpL3‐AU1235, (b) MmpL3‐ICA38, (c) MmpL3‐NITD349, (d) MmpL3‐Rimonabant, (e) MmpL3‐SPIRO, (f) MmpL3‐SQ109, and (g) MmpL3‐ST004. Residues contributing more than 1 kcal/mol to the free energy in at least one trajectory are reported.

2.4. Inhibitors suppress conformation transitions of the proton channel opening and closing

Based on the preceding analysis, inhibitors bind to the central channel within the TMD, interacting with ASP251 or ASP640 through hydrogen bonds. To characterize the impact of inhibitor binding on the central channel of TMD, we defined and calculated the proton channel size (distance between ASP251 on TM4 and TYR641 on TM10 as illustrated in Figure 4a). Figure 4b shows a clear difference in the proton channel size between inhibitor‐free and inhibitor‐bound systems. In the MmpL3‐APO system, the proton channel size predominantly ranges from 9 Å to 12 Å. However, in inhibitor‐bound systems, the proton channel size is generally larger, mainly ranging from 10 Å to 13 Å, with only the MmpL3‐AU1235 system showing a few values between 9 Å and 10 Å. Experimentally determined apo forms structures mostly exhibit sizes less than 11 Å, except for the 7K8D structure. In contrast, the inhibitor‐bound structures typically display a larger proton channel size, with exceeding 11 Å (refer to Table S5). Therefore, in the apo state, the size of the proton channel exhibits considerable variability, with the majority predominantly in a closed state but capable of transitioning to an open state. The binding of inhibitors may directly influence the size of the proton channel, locking it into an open state and preventing the transition from the open state to a closed one.

FIGURE 4.

The distribution of proton channel size in all simulation systems. (a) Definition of the proton channel size (PDB ID: 7NVH). The distance is depicted in the figure, with units in angstroms (Å). (b) The distribution of the proton channel size in the apo form and inhibitor‐bound forms of Mtb MmpL3 systems.

2.5. Difference in free energy landscape of the TMD size and PD size in MmpL3‐APO and inhibitor‐bound systems

To analyze the conformations of the TMD and PD, we selected the TMD size (distance between THR176 on the TM2 and ALA557 on the TM8) (Su et al., 2021) and the PD size (distance between VAL91 on the PN and SER474 on the PC) to construct the free energy landscape, as shown in Figure 5a. Figure 5b illustrates that in all simulated systems, the distribution range of the TMD size is relatively consistent, while the PD size in the MmpL3‐APO system is slightly larger than that in the inhibitor‐bound systems. Moreover, in the MmpL3‐APO system, there are two primary conformation states. The first conformation state showcases a PD size of approximately 43 Å with a corresponding TMD size around 31.5 Å at the core position. The second conformation state displays a PD size of approximately 39 Å with a TMD size around 33 Å. In the former state, the PD size is larger and the TMD size slightly smaller, whereas in the latter state, the PD size is smaller and the TMD size slightly larger. Su et al. (Su et al., 2021), based on structural biology, demonstrated that the primary conformational change in M. smegmatis MmpL3 involves a rigid body rotational motion of the PC relative to the PN subdomain. This motion imparts plasticity to the PD cavity, allowing the cavity volume to adjust to accommodate lipid binding. Additionally, the expansion of the PD cavity is associated with conformational changes in several transmembrane helices, which accommodate the rearrangement of the PD. In our study, we similarly observed the opening and closing motion of the PD. Interestingly, our study revealed an inverse correlation in the opening and closing dynamics of the PD and TMD.

FIGURE 5.

The free energy landscape for the TMD size and PD size for all simulation systems. (a) Definition of the PD size and TMD size (PDB ID: 7NVH). The distance is depicted in the figure, with units in angstroms (Å). (b) Free energy landscapes depicting the apo form and inhibitor‐bound forms of Mtb MmpL3, relative to variations in the TMD size and PD size as determined by MD simulations. Magenta triangle symbols denote experimentally resolved structures of the Mtb apo form of MmpL3 (PDB ID: 7NVH), solid black circles and cyan triangle symbols represent experimentally resolved structures of the apo form and inhibitor‐bound forms of M. smegmatis MmpL3, respectively.

However, in inhibitor‐bound systems, the conformations with low energy are mostly distributed within the same TMD size range. It shows no evidence of anticorrelated movement between the TMD and the PD. Furthermore, the majority of experimental structures align with the regions of larger PD size, with no structures found in the regions of smaller PD size, potentially indicating that the latter conformations represent intermediate states of conformation changes.

2.6. Two‐state conformation transition of the TMM binding site in MmpL3‐APO system is inhibited in inhibitor‐bound systems

It is believed that TMM is transported from the TMM pocket to the PD cavity through TMM entry (Su et al., 2021). To characterize the conformations at the TMM binding pocket within the TMD, the TMM pocket size (distance between PRO402 and VAL572) and the TMM entry size (distance between SER418 and ALA519) were selected to construct the free energy landscape. Figure 6a illustrates the differences in the TMM binding site between MmpL3‐APO and inhibitor‐bound systems. In the MmpL3‐APO system, it depicts two distinct local minima, indicating the potential of a two‐state conformation transition. In the state with denser distribution, the TMM entry measures approximately 11 Å, and the TMM pocket is about 30.5 Å, indicative of a closed‐like state of the TMM pocket. Conversely, in the sparser distribution state, the dimensions are about 9 Å for the TMM entry and 33 Å for the TMM pocket, suggesting a more open TMM pocket and a relatively closed TMM entry (as shown in Figure 6b). This conformation implies that constriction at the TMM entry may facilitate the expansion of the TMM pocket. Experimentally resolved three‐dimensional structures predominantly cluster near the denser distribution, indicating a preference for the closed‐like state of the TMM pocket. The open TMM pocket and closed TMM entry conformation, observed in the sparser distribution, likely represent intermediate states that facilitate TMM binding from the IM to the site. Su et al. (Su et al., 2021) proposed that the flexible loops composed of residues R523‐D525 and S423‐G426 of M. smegmatis MmpL3 form a gate that regulates the opening and closing of the TMM entry, which aligns well with our observations. Additionally, the structure of the MmpL3‐TMM complex indicates that TM8 and TM9 play a critical role in anchoring the hydrophobic tail of the bound TMM. The subtle motions of the V‐shaped structure formed by these two transmembrane helices might facilitate the flipping of the TMM molecule within the cytoplasmic membrane. In our study, we found that the TMM pocket formed by TM7 and TM8 undergoes conformational changes resembling opening and closing motions, which might control the initial lipid binding to MmpL3, subsequently triggering the flipping and transport process.

FIGURE 6.

The conformation distribution of the TMM binding pocket. (a) Free energy landscapes depicting the apo form and inhibitor‐bound forms of Mtb MmpL3, relative to variations in the TMM pocket size and TMM entry size as determined by MD simulations. (b) The apo forms of Mtb MmpL3 from PDB ID 7NVH (cyan) and our MD simulations from the M0 state (magenta). The region between TM7 and TM8 constitutes the TMM pocket, with the distance between SER418 (blue) and ALA519 (blue) representing the TMM entry size. Magenta triangle symbols denote experimentally resolved structures of the Mtb apo form of MmpL3 (PDB ID: 7NVH), solid black circles and cyan triangle symbols represent experimentally resolved structures of the apo form and inhibitor‐bound forms of M. smegmatis MmpL3, respectively.

However, inhibitor binding significantly alters the free energy landscape, as illustrated in Figure 6a. MmpL3‐AU1235, though conformational similar to the MmpL3‐APO, displays a broader distribution and lacks two‐state transition. The changes observed in ICA38, NITD349, and Rimonabant suggest the stabilization of alternative conformational states distinct from those in MmpL3‐APO. Moreover, MmpL3‐SPIRO and MmpL3‐SQ109 exhibit enlarged TMM pocket sizes without corresponding reduction in TMM entry size. Therefore, inhibitor binding diversifies the conformation landscape at the TMM binding site and potentially decouples the coordinated movements of the TMM entry and the TMM pocket, affecting TMM's binding and translocation within the TMD.

2.7. Coupled motions in MmpL3‐APO system and inhibitor‐bound systems

The conformation analysis highlights the coupled motions between PD and TMD domains, TMM binding pocket, and TMM entry in MmpL3‐APO system, leading us to conduct a thorough investigation into these motions. We utilized the DCCM analysis, focusing on the Cα atoms of the whole protein, to observe these coupled motions in Mtb MmpL3. It reveals synergistic movements within the PN and PC subdomains, with or without inhibitors, as depicted in the black box in Figure 7. Notably, the PN and PC subdomains exhibit opposing motion patterns, as highlighted in the green box in Figure 7, indicating the dynamic opening and closing of the central cavity within the PD. This observation is consistent with changes in the PD size (Figure 5).

FIGURE 7.

The DCCM analysis focused on the Cα atoms in MD simulations for both the apo and inhibitor‐bound forms Mtb MmpL3.

Following the analysis within the PD, we examined the motion patterns between the TMD and the PD. In the MmpL3‐APO system, opposing motion patterns are identified between the PN subdomain and the primary helices of TMD1 (TM1–TM6). In contrast, synergistic motion patterns are observed between the PN subdomain and TMD2 (TM7‐TM12), as highlighted in the red box in Figure 7. Concurrently, the PC subdomain and the motion states of TMD1 and TMD2 appeared to counter the PN subdomain, as depicted in the cyan box in Figure 7. This suggests anti‐correlated movement between the TMD and the PD, aligning closely with the distribution shown in Figure 5. Similarly, Su et al. proposed that the conformational change of the PC is coupled with the displacement of the transmembrane helices TM7–TM10 (Su et al., 2021). However, inhibitor‐bound systems exhibited variability in the motion patterns between the TMD and the PD (Figure 7).

The PCA analysis, focusing on the Cα atoms, was performed to identify the primary motion modes. The first five motion modes in PCA collectively accounted for approximately 80% of the total variance (Figure S12), which were selected for visualization using the VMD software (Humphrey et al., 1996) and the NMWiz plugin (Bakan et al., 2011). In the MmpL3‐APO system, an intriguing observation is the coupled motions among the PD, the TMM pocket, and the TMM entry. In the first motion mode, the opening of PD coincides with the opening of the TMM entry, while conversely, the TMM pocket transitions to a closed state (Figure S13 and Video S1). This coupled motion results in the two‐state transition observed in Figure 6. However, these coupled motions are not observed in the inhibitor‐bound systems. Furthermore, in both inhibitor‐free and inhibitor‐bound systems, the primary motion mode involves opposing directional movements of the PN and PC subdomains (Figure S13), which is in line with the findings of the DCCM analysis.

2.8. MSM reveals conformation transition in MmpL3‐APO system

To further investigate the overall conformation states and transitions in MmpL3‐APO system, an MSM was constructed based on MD simulation trajectories, using the PyEMMA 2.5.7 package (Scherer et al., 2015). Specific residue pairs were selected for distance calculations between Cα atoms as primary features to accurately describe the conformation states, as depicted in Figures 4a, 5a, and 6b. These include the size of the proton channel (ASP251‐TYR641), TYR252‐ASP640, the TMD size (THR176‐ALA557), the PD size (VAL91‐SER474), the TMM entry size (SER418‐ALA519), and the TMM pocket size (PRO402‐VAL572). These features were subsequently projected into three dimensions using the time‐lagged independent component analysis (tICA) method (Molgedey & Schuster, 1994; Pérez‐Hernández et al., ²⁰¹³) (Figure S14A). Using the k‐means algorithm, the conformations were clustered into 200 microstates. The analysis of the relationship between lag time (τ) and relaxation time, depicted in Figure S14B, revealed that within the 300–500 ps range, the relaxation time scale remained nearly constant, leading to the selection of τ as 400 ps. The Chapman‐Kolmogorov (CK) test method was employed to validate the appropriateness of the chosen parameters (Figure S14C). The 200 clustered microstates were further grouped into four metastable states using the Perron Cluster Cluster Analysis (PCCA++) (Röblitz & Weber, 2013). In addition, the mean first passage time (MFPT) between different metastable states was computed, and the conformation parameters of different macrostates were calculated from 500 conformations.

The MSM identifies four macrostates, designated as “S1”–“S4” states (see Figure 8a,b). “S1” and “S3” account for 30.08% and 33.64% of the distribution, respectively. Meanwhile, “S2” and “S4” represent smaller conformation proportions, at 13.76% and 20.95%, respectively. Within “S1” and “S2” states, the proton channel exhibits smaller average dimensions of 9.55 Å and 10.08 Å, respectively, with the PD size of “S1” (43.26 Å) being larger than that of “S2” (39.23 Å). With respect to the TMM pocket, “S1” (30.47 Å) is smaller than “S2” (32.84 Å), while the TMM entry in “S1” (10.95 Å) is larger than in “S2” (9.25 Å). The proton channel in “S3” and “S4” states is larger than that in “S1” and “S2” states, measuring 10.80 Å and 10.67 Å, respectively. “S4” state displays a larger PD size than “S3” state, whereas the TMM pocket size is slightly smaller than that in “S3” state. In terms of the TMM entry, “S3” and “S4” exhibit comparable values, measuring 11.18 Å and 11.41 Å, respectively.

FIGURE 8.

The MSM analysis for MmpL3‐APO system. (a) Free energy landscape for the first two components of tICA in MSM analysis, where different circles roughly denote distinct states. Black circles represent the proton channel size that is larger than that represented by red circles. (b) Analysis of conformation kinetics among various macrostates using MFPT. (c) The representative structures extracted from “S1” to “S4” states. Arrows indicate the MFPT in ns, and numbers within the boxes denote the conformation parameters.

The MFPTs between the four macrostates are on the order of 10s of nanoseconds (ns) as shown in Figure 8b. During the transition between “S1” and “S2,” simultaneous opening and closing motions of the PD and TMM entry are observed. At the same time, the TMM pocket exhibits opposing movements. The proton channel remains relatively constant throughout this transition (as shown in Figure 8b,c). The transitions between them are relatively slow, with transition times of 36.3 ± 2.0 ns and 96.1 ± 6.3 ns, indicating a considerable energy barrier. In the transition between “S2” and “S3” states, there are opposing transitions between the open and closed states of the TMM pocket and TMM entry, accompanied by a minor change in PD size. Additionally, the “S3” state exhibits an increase in proton channel size compared to the “S2” state. The transition time from “S2” to “S3” at 10.8 ± 0.7 ns is substantially quicker than its reverse at 87.0 ± 6.1 ns, highlighting a preferential stability toward the “S3” state. In the transition from states “S3” to “S4,” the proton channel and TMM entry maintain a relatively open state, whereas the TMM pocket remains relatively closed. Simultaneously, the increase in PD size suggests the gradual opening of the PD cavity. In particular, the transition between states “S3” and “S4” is the most rapid, with forward and backward transition times of 9.7 ± 0.4 ns and 4.8 ± 0.1 ns, respectively. In states “S1” and “S4,” changes in PD size and TMM pocket size are negligible, but alterations are noted in the proton channel (Figure S15A). The transitions between “S1” and “S4” states show a moderate pace, with forward and backward times of 15.6 ± 0.5 ns and 23.7 ± 1.0 ns, respectively, indicating a balanced dynamic between these states.

In summary, within the MmpL3‐APO system, when the proton channel is narrow, the closure of the PD leads to the closure of the TMM entry, while the TMM pocket opens. With a wider proton channel, the TMM pocket remains closed while the TMM entry is open, with the PD rapidly transitioning between open and closed states. Consequently, there exists coupled motion among functional substructures in the MmpL3‐APO system.

2.9. Proposed TMM transport mechanism by Mtb MmpL3 and inhibition mechanism by inhibitors

TMM may undergo a two‐step flipping mechanism across the IM mediated by MmpL3. The first step relies on the PMF to translocate TMM from the inner to the outer leaflet of the IM (Xu et al., 2017). The second step involves MmpL3‐mediated transport of TMM from the outer leaflet of the IM to the PD (Su et al., 2021). This study uncovers various conformations of the apo form of Mtb MmpL3 and the transitions between these conformations, providing more structural details at the molecular level. Importantly, MD simulations may reveal the coupling between proton relay and TMM transport during the second step, a dynamic process that cannot be captured by static structures. In this study, the proton channel can be delineated into two distinct regions based on their size: one smaller and the other larger. Experimentally resolved structures of the apo form of MmpL3 predominantly occupy the smaller region. In contrast, MmpL3 bound with inhibitors mainly populates the larger region, as listed in Table S5. In AcrB, one of the most extensively studied RND transporters, the protonation of ASP407 and ASP408 moves them away from LYS544, causing a repulsion effect that induces a minor conformational change in the TMD (Eicher et al., 2014). This is similar to the observed changes in the proton channel that we simulated. Thus, we infer that protons passage through the proton channel leads to the protonation of ASP251 and ASP640, expanding the size of the proton channel and further characterizing PMF during TMM transport. The experimentally resolved structure of M. smegmatis MmpL3 (PDB ID: 7N6B) reveals that TMM binds within the TMM pocket and the PD cavity. Our results show that the TMM pocket undergoes conformation transitions between relatively open and closed states, with the open conformation occurring when both the PD and the proton channel exhibit smaller sizes (Figure 8c). Therefore, we infer that the relatively open conformation of the TMM pocket may facilitate the binding of TMM from the outer leaflet of the IM. Throughout this process, the proton channel remains relatively narrow, suggesting that TMM binding could be spontaneous and independent of the PMF. This hypothesis is supported by the work reported by Li et al. (2023). In addition, in the larger proton channel (PMF may exist), the PD size undergoes rapid transitions between open and closed conformations, with relatively open states for the TMM entry (Figure 8b,c). This may signify the translocation of TMM into the PD cavity. Therefore, PMF may be related to the translocation of TMM from the TMM pocket to the PD cavity, but not to the initial binding of TMM to the TMM pocket. Moreover, the TMD exhibits a movement pattern opposite to the PD, with the PD displaying a wider range of opening and closing movements, while the TMD undergoes relatively minor changes. This differential movement pattern may be related to the release of TMM from the PD cavity. It is hypothesized that the PD plays a role in handing off TMM to an unidentified chaperone system to transport TMM to the OM, as discussed in the review (Williams & Abramovitch, 2023).

Based on the above analysis, a potential mechanism for TMM transport by Mtb MmpL3 is proposed, as illustrated in Figure 9: in the “S1” state, one TMM is handed off to an unidentified chaperone system by the PD, and another TMM begins to bind to the TMD. In the “S2” state, the TMM pocket opens, facilitating TMM binding. In the “S3” state, when the TMM pocket begins to close and the TMM entry begins to open, the TMM starts to translocate into the PD cavity. In the “S4” state, the TMM binds to the PD cavity. Su et al. (Su et al., 2021) proposed a TMM transport mechanism wherein TMM flips from the inner leaflet to the outer leaflet, binds between TM7 and TM8, and then enters the PD cavity. Subsequently, TMM migrates between the PN and PC regions before finally releasing into the periplasm. They proposed that PC may undergo rigid body rotational motion relative to the PN subdomain, accompanied by displacement of the transmembrane helices TM7–TM10. These two primary conformational changes are coupled and may facilitate lipid translocation. Additionally, the narrowest region of the elongated channel is surrounded by a flexible loop composed of residues R523–D525 on one side, and residues S423–G426 on the other. These residues likely control the TMM entry opening and closing through coupled dynamics with PC, thereby promoting transport cycling. This closely resembles the TMM translocation pathway we proposed. Our observations reveal additional coupled motions spanning the proton channel, TMM pocket, TMM entry, and PD, providing further details and proposing a potential mechanism by which proton relay may play a role in TMM translocation. In addition, Carbone et al. suggest that the conformation changes of M. smegmatis MmpL3 are coupled with the TMM transport process. The study posits that the open‐close motion of the TMDs, likely driven by proton translocation, induces the open‐close motion of the PDs, facilitating TMM transport (Carbone et al., 2023). This bears some similarity to our observations, where we found significant conformational changes in the PD, while only minor conformational changes occurred in the TMD. We believe that the proton relay may directly influence the size of the proton channel, thereby further affecting the dynamic structural properties of MmpL3 and, consequently, its function.

FIGURE 9.

Proposed mechanism of TMM transportation by Mtb MmpL3 and inhibition mechanism by inhibitors. Ovals filled with cornflower blue and orange represent the PN and PC subdomains, respectively. Rectangles filled with cornflower blue and orange represent the TMD1 and TMD2 subdomains, respectively. Rectangles filled with pink, yellow, blue, light green, and steel blue represent TM2, TM4, TM7, TM8, and TM10, respectively. An oval with two line segments filled with purple represents TMM. Arrows indicate the TMM translocation process, with red arrows representing the steps potentially blocked by inhibitors. The red dashed box delineates the conformation states of MmpL3 that are absent following inhibitor binding. The unidentified chaperone system is not labeled in the figure.

Simulations with inhibitor‐bound systems reveal distinct conformation distributions compared to the MmpL3‐APO system. The proton channel consistently exhibits an enlarged size, as depicted in Figure 4. This enlargement of the proton channel results in the absence of specific conformations within the red dashed box in Figure 9, which may be pivotal for TMM translocation. Specifically, the opening of the TMM pocket coincides with the closure of the TMM entry, while the proton channel remains closed. This conformation likely facilitates TMM binding to the TMD, but the presence of inhibitors modifies this state (Figure 6). Additionally, the reverse motion between the TMD and PD could facilitate the opening of the PD cavity, thereby aiding TMM release. However, the binding of inhibitors appears to impede this reverse motion (Figure 5). In summary, inhibitors may disrupt the TMM translocation process by directly binding to the central channel of the TMD, increasing the size of the proton channel, and thereby disrupting the coupled motions of MmpL3.

3. CONCLUSION

In this study, we utilized MD simulations to investigate the binding modes between inhibitors and MmpL3, as well as the transport and inhibition mechanisms of TMM. MD simulations of the complex reveal that inhibitors with different scaffold structures bind stably to the central channel of the TMD. Free energy calculations indicate that non‐polar interactions predominantly contributed to the binding energy. Hydrogen bond analysis demonstrates that the inhibitors in this work primarily form hydrogen bonds with either ASP251, ASP640, or both. Furthermore, DCCM and PCA analyses identify several coupled motions among functional substructures in the apo form of MmpL3, with distinct conformational states delineated through MSM analysis. These coupled motions and varied conformational states are believed to facilitate TMM transport. Specifically, when the proton channel narrows, the closure of PD leads to the opening of the TMM pocket, potentially facilitating TMM binding within the pocket. Conversely, when the proton channel enlarges, the PD cavity opens promptly, maintaining an open state for the TMM entry. This transition likely correlates with TMM translocation into the PD cavity and subsequent release. Based on these observations, a proposed transport mechanism for TMM has been formulated. However, coupled motions among the domains are absent in simulations of inhibitor‐bound forms of MmpL3. Consequently, inhibitors may obstruct the closure of the proton channel, impeding the coordinated movement of MmpL3 and indirectly inhibiting TMM translocation. In conclusion, this study employs computational biology methods to elucidate the TMM transport and inhibition mechanisms of MmpL3 at the molecular and atomic levels. It offers a new perspective on understanding the structure and function of MmpL3 and lays a theoretical foundation for developing anti‐TB drugs targeting Mtb MmpL3. Further investigations into the protonation effects of ASP251 and ASP640 on MmpL3's structure will contribute to a more comprehensive understanding of its functionality.

4. MATERIALS AND METHODS

4.1. Systems preparation

The homology modeling process encompassed several crucial steps. Initially, the apo and inhibitor‐bound forms of M. smegmatis MmpL3 underwent energy optimization and preparation using the QuickPrep module within the Molecular Operating Environment (MOE) software (MOE, ²⁰²⁴). Subsequently, homology modeling and structure completion were carried out by means of the Protein Structure Prediction Server Robetta (https://robetta.bakerlab.org/) (Kim et al., 2004). The protein sequence for Mtb MmpL3 (residues 1–730) was retrieved from the UniProt database with Uniprot ID P9WJV5 (Consortium, 2018). For the homology modeling of the apo form of Mtb MmpL3, the X‐ray crystallographic protein structure (PDB ID: 7K7M) was employed as a template (Su et al., 2021). Regarding the homology modeling of the seven inhibitor‐bound forms of Mtb MmpL3, two distinct templates were utilized. Template 1 employed different inhibitor‐bound structures of M. smegmatis MmpL3, as depicted in Table S1. Meanwhile, template 2 incorporated the X‐ray crystallographic protein structures of the apo form of M. smegmatis MmpL3 (PDB ID: 7K7M) to construct the missing residues 343–377 between TMD1 and TMD2. To generate complexes of Mtb MmpL3 with various inhibitors, the homology‐modeled structure of Mtb MmpL3 was superimposed onto the inhibitor‐bound template M. smegmatis MmpL3. Then, the inhibitors from the M. smegmatis MmpL3 structure were mapped onto the Mtb MmpL3 protein. Following this, the resulting complex of Mtb MmpL3 and the inhibitor underwent energy optimization using the QuickPrep module in the MOE software. A total of eight systems were constructed (Table S2).

The eight structures, as previously outlined, were each incorporated into separate lipid bilayers composed of POPC and POPE in a 1:1 molecular ratio. The dimensions of the water box varied slightly, as detailed in Table S2, and the configuration was established using the CHARMM‐GUI Membrane Builder web server (http://www.charmm-gui.org/) (Jo et al., 2008). All systems were supplemented with 150 mM NaCl and additional neutralizing counterions.

Parameters for various inhibitors were generated using the Gaussian 09 software (Frisch et al., 2009) and the Antechamber module of AmberTools22 (Wang et al., 2000). Following this, topology and coordinate files for the complexes were produced. Both SPIRO and ST004 possess a single positive charge, whereas SQ109 features a diprotonated ethylenediamine unit, as depicted in Figure S16. The other small molecules within these systems are neutral. The electrostatic potentials of the inhibitors were calculated at the Hartree‐Fock level using a 6‐31G* basis set, with partial charges on atoms determined by the RESP charge‐fitting method (Breneman & Wiberg, 1990). The Lipid 17 force field (Gould et al., 2018) was utilized for lipid molecules, and the General Amber Force Field 2 (GAFF2) (He et al., 2020; Wang et al., ²⁰⁰⁴) was used for the inhibitors. Proteins were modeled using the ff19SB force field (Tian et al., 2020), and missing hydrogen atoms were added via the tleap tool in Amber 22 (Case et al., 2022). Rectangular water boxes, filled with TIP3P water molecules (Jorgensen et al., 1983), were constructed, leading to each simulation system containing more than 170,000 atoms, as outlined in Table S2.

4.2. MD simulation setup

All MD simulations in this study were conducted using the Amber 22 package (Case et al., 2022). A systematic and multi‐stage relaxation strategy was employed to refine the systems and alleviate any unreasonable atomic collisions. The process began with a three‐stage energy minimization protocol. In the first stage, constraints were applied to both the membrane and the complex, followed by applying constraints solely to the complex in the second stage. The third stage allowed for simulations under unrestricted conditions, with each stage running for 5000 steps to ensure proper adjustment. This was followed by a heating phase, where the temperature was gradually increased from 0 to 303 K. Throughout this phase, harmonic constraints were maintained on the membrane and complex in the NVT ensemble, over a simulation duration of 100 ps. During the equilibration phase in the NPT ensemble, constraints on the membrane and complex were systematically reduced across four stages. The force constants employed were 10, 5, 2, and 1 kcal/(mol·Å²), respectively, with each stage lasting 100 ps. A final equilibrium simulation of 200 ps was conducted, with adjustments made to the solvent density.

In the production simulations conducted under the NPT ensemble, multiple parallel simulations were performed for each system, as detailed in Table S2. To ensure numerical stability and mitigate rapid fluctuations in the system, constraints were applied to all bonds including hydrogen bonds, using the SHAKE algorithm (Miyamoto & Kollman, 1992). The Particle Mesh Ewald method (Toukmaji et al., 2000) was employed for electrostatic interaction calculations, with a short‐range cutoff set to 10.0 Å. Simulations were carried out under constant pressure conditions, maintaining a reference pressure of 1.0 bar and using a pressure coupling constant of 1.0 ps. Non‐bonded interactions were truncated at a distance of 10.0 Å. Temperature control was achieved using the Nosè‐Hoover thermostat, with a coupling constant of 2.0 ps⁻¹. The simulation time step was set at 2 fs, and trajectory data were recorded every 5 ps. The cumulative duration of all simulation trajectories exceeded eight microseconds ($\mu$s).

4.3. MM/GBSA calculation

To assess the binding affinity of various inhibitors with Mtb MmpL3, the MM/GBSA method was utilized to estimate the binding free energy between the protein and inhibitors. This approach has proven successful in numerous studies (Zhang et al., 2020; Zhang, An, et al., ²⁰¹⁹). The binding free energy was computed using the following equation:

$$\Delta G_{\text{bind}}=G_{\text{complex}}-G_{\text{receptor}}-G_{\text{ligand}}$$

Here, $G_{\text{complex}}$, $G_{\text{receptor}}$, and $G_{\text{ligand}}$ represent the free energy associated with the complex, protein, and ligands, respectively. The free energy for each molecular entity was determined by averaging across snapshots extracted from multiple trajectories. The aforementioned quantities can be calculated using the following formulas:

$$G=E_{\mathrm{MM}}+G_{\mathrm{sol}}-\mathit{TS}$$

$$E_{\mathrm{MM}}=E_{\mathrm{int}}+E_{\mathrm{ele}}+E_{\mathrm{vdw}}$$

$$E_{\mathrm{int}}=E_{\text{bond}}+E_{\text{angle}}+E_{\text{torsion}}$$

$$G_{\mathrm{sol}}=G_{\mathrm{GB}}+G_{\mathrm{SA}}$$

$$G_{\mathrm{SA}}=\gamma \times \text{SASA}+\beta$$

In the above description, $E_{\mathrm{MM}}$ refers to the gas‐phase energy. $E_{\mathrm{int}}$ represents internal energy, encompassing bond energy ($E_{\text{bond}}$), angle energy ($E_{\text{angle}}$), and torsional energy ($E_{\text{torsion}}$). $E_{\mathrm{ele}}$ and $E_{\mathrm{vdw}}$ correspond to coulombic energy and Van der Waals energy, respectively. $G_{\mathrm{sol}}$ denotes solvation‐free energy, which can be decomposed into polar solvation‐free energy ($G_{\mathrm{GB}}$) and non‐polar solvation free energy ($G_{\mathrm{SA}}$). $G_{\mathrm{GB}}$ is computed by solving the GB equation, with dielectric constants set to 1.0 and 80.0 for solute and solvent, respectively. $G_{\mathrm{SA}}$ is estimated by solvent‐accessible surface area using a water probe radius of 1.4 Å. The surface tension constant $\gamma$ is set to 0.0072 kcal/(mol·Å²), and the non‐polar solvation‐free energy term $\beta$ is set to 0. T and S represent temperature and total solute entropy. The contribution of entropy can be estimated through normal mode analysis (NMA). To save the computational resource, the truncated structures can be used to calculate the NMA conformation entropies (Genheden et al., 2012; Sun et al., ²⁰¹⁸).

4.4. Markov state model construction and validation

The primary objective of MSM analysis is to derive long‐term kinetic information from short‐duration MD simulation trajectories (Husic & Pande, 2018; Wang et al., ²⁰¹⁸). This approach enables the capture of slow structural transitions within the conformation ensemble generated by MD trajectories. Conformation changes play a crucial role in the functional behavior of biomolecules. These rare events can be identified by constructing a transition probability matrix over a discretized conformation state space (Pérez‐Hernández et al., 2013; Prinz et al., ²⁰¹¹). Typically, a process is deemed Markovian when the model adheres to the memoryless master equation:

$$P\left(n\mathrm{\tau }\right)={\left[T\left(n\mathrm{\tau }\right)\right]}^{\mathrm{n}}P\left(0\right)$$

In this equation, P represents the state population vector, T denotes the transition probability matrix, and τ refers to the model's lag time.

For a specified lag time τ, the implied timescale can be determined by the following method:

$$t_i=\frac{-\mathrm{\tau }}{\mathit{ln}\left[{\lambda }_i\left(\mathrm{\tau }\right)\right]}$$

In this context, $t_{\mathrm{i}}$ represents the implied timescale, and ${\lambda }_{\mathrm{i}}$ denotes an eigenvalue of the transition matrix T (t). When the implied timescales $t_{\mathrm{i}}$ become roughly constant as the lag time τ increases, it indicates that the model is beginning to fulfill the Markov property.

The CK test (Noé et al., 2009; Prinz et al., ²⁰¹¹) calculates transition probabilities between meta‐stable states at different lag times. This test assesses the degree to which the system's behavior is Markovian, meaning that transitions between microstates depend solely on the current state at time t and are independent of the historical sequence of state transitions.

$$\left[T\left(n\mathrm{\tau }\right)\right]\approx {\left[T\left(\mathrm{\tau }\right)\right]}^{\mathrm{n}}$$

In this context, T ($\text{τ}$) represents the transition matrix derived from the data at a lag time of τ, with n being an integer that signifies the number of steps involved.

AUTHOR CONTRIBUTIONS

Likun Zhao: Conceptualization; methodology; validation; visualization; investigation; formal analysis; software; writing – review and editing; writing – original draft; data curation. Bo Liu: Investigation; methodology; software; visualization. Henry H. Y. Tong: Methodology; supervision. Xiaojun Yao: Methodology. Huanxiang Liu: Methodology; investigation; funding acquisition; resources. Qianqian Zhang: Methodology; supervision; funding acquisition; resources; project administration; writing – review and editing; formal analysis; data curation; investigation; conceptualization.

FUNDING INFORMATION

This work was funded by Macao Polytechnic University (Grant No. RP/FCA‐11/2023) and the joint project of The Science and Technology Development of Macau and the National Natural Science Foundation of China (FDCT‐NSFC) (Grant No. 0043/2023/AFJ).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Data S1. Supporting Information.

Video S1. Animation of the first motion mode of the PCA analysis for MmpL3‐APO system.

Zhao L, Liu B, Tong HHY, Yao X, Liu H, Zhang Q. Inhibitor binding and disruption of coupled motions in MmpL3 protein: Unraveling the mechanism of trehalose monomycolate transport. Protein Science. 2024;33(10):e5166. 10.1002/pro.5166