1. Structural Evaluation of EtfD: AlphaFold provides pLDDT and PAE metrics to assess predicted model's accuracy. The pLDDT score, ranging from 0 to 100, estimates the agreement between predicted and experimental protein structures, effectively assessing local model quality, with lower scores typically correlating with a higher likelihood of intrinsic disorder [30]. The predicted model of the EtfD has an average pLDDT value of 82.56, indicating overall high confidence, with specific regions exceeding 90. However, regions 342 to 357 and 745 to 882 display low values below 50, suggesting likely disorder (Fig. 2a, b). PAE indicates the predicted error between relative positions of residue pairs in protein structures, with low values suggesting well-defined relationships and high values indicating unreliable positions [45]. The PAE 2D heatmap shows low confidence in the positions of all residues relative to regions 342-357 and 745-882, which also have low pLDDT values, lack secondary structure, and exhibit coiled, ribbon-like appearance, predicting disorder (Fig. 2d). The flDPnn, a disorder prediction tool utilizing deep neural networks, accurately predicts disorder and fully disordered proteins, while also generating putative functions for predicted intrinsically disordered regions (IDRs) [33]. The flDPnn server identified regions as IDRs that corresponded with low pLDDT and high PAE scores from AlphaFold. However, it also predicted these regions to have putative functions in binding with various biomolecules (Fig. 2c). MolProbity is a tool used to validate and analyze biomolecular 3D structures by assessing factors like geometry, steric clashes, and overall structural validity [32]. MolProbity analysis of EtfD revealed 93% of residues in the favored region and additional 4.6% in the allowed region of the Ramachandran plot, with 21 outliers observed (Fig. S1). Notably, 20 out of 21 outliers corresponded to residues within predicted disordered regions.
The overall confidence in EtfD's predicted structure is high, except for two disordered regions with the lowest pLDDT scores, which also exhibit uncertain positions relative to the protein, as indicated by high PAE values. However, the rest of the structure shows high reliability, with many regions having pLDDT values over 90, comparable to experimental structures. This accuracy supports various applications, especially for precise tasks like characterizing binding sites. The connecting loops also have favorable pLDDT values (70-90), indicating good backbone prediction. This assessment confirms the model's reliability, making it suitable for further research, especially in drug design against this protein.
2. Binding Site Prediction and Characterization: Identifying a protein's ligand binding site is essential for understanding its function and designing therapeutic compounds to modulate its activity [46]. The COACH metaserver detected several ligand binding sites, with the top site having the highest confidence score of 0.20, selected for further analysis (Fig. 3a). The figure (Fig. 3b) depicts ligand binding residues identified by the COACH, along with their corresponding residue-specific probabilities. The predicted ligand, Iron Sulfur Cluster (SF4), directly interacts with residues THR 296, CYS 298, GLY 299, CYS 301, CYS 402, and HIS 408. Although not predicted as a binding site residue, HIS 408 plays a crucial role, engaging in two pi-sulfur interactions and a hydrogen bond with two sulfur atoms of SF4. Additional significant interactions of SF4 include hydrogen bonds with CYS 301, GLY 299, CYS 298, and two with THR 296. SF4 forms coordinate bonds with CYS 295, CYS 298, and CYS 301, with the fourth iron atom participating in a metal acceptor interaction with CYS 402 (Fig. 3c, d).
Examining the intricate interactions between binding site and its surroundings is crucial for understanding structural stability and functional significance. Binding site residues engage in diverse interactions with surrounding residues, encompassing hydrogen bonds, electrostatic interactions, alkyl interactions, and salt bridges (Fig. 3e). Notably, Glutamate 297, Glycine 299, Lysine 317, and Cysteine 402 are involved in multiple hydrogen bonds with surrounding residues. The identification of the predicted binding site in EtfD, along with its probable ligand, an iron sulfur cluster, reinforces our hypothesis regarding the protein's metabolic involvement. Targeting this site offers a promising approach for designing drugs against Mtb, signifying the significance of this structural insight for therapy development.
3. Virtual Screening: A database of 3,447 approved drug molecules was screened against EtfD using PyRx 0.8. From Vina docking results, the top 20 compounds with binding affinities greater than -8.7 kcal/mol were prioritized for repurposing ( Table S1).
4. Molecular Docking: The top 20 molecules were re-docked against EtfD using AD4, with binding energies ranging from -7.31 to -10.47 kcal/mol (Table 1). The top hits for further MD simulation studies are ZINC000003812842 (diacerein), ZINC000000603195 (levonadifloxacin), and ZINC000003607120 (gatifloxacin), with binding energies of -10.47, -10.04, and -10.02 kcal/mol, respectively. Diacerein forms hydrogen bonds with ARG133, ARG146, and ARG300, and a carbon hydrogen bond with ALA155. It also exhibits hydrophobic interactions with HIS235 and CYS298, and electrostatic interactions with ARG146. Levonadifloxacin interacts with EtfD through hydrogen bonds involving ARG133, ARG146, ARG300, and GLY154, and a carbon hydrogen bond with ALA155. Hydrophobic interactions, including pi-alkyl and alkyl interactions, occur with HIS235, CYS298, and HIS74, enhancing stability. Gatifloxacin predominantly relies on hydrophobic interactions, involving key amino acids PHE129, ILE132, ARG133, PHE147, ALA155, VAL158, and LYS232. PHE129 forms four significant hydrophobic interactions. Additionally, ARG133 and ARG300 contribute to hydrogen bonding, along with ALA155. HIS74 and HIS235 facilitate halogen (fluorine) interactions, with HIS74 also engaging in a hydrogen bond. ARG146 stabilizes the interaction through electrostatic interactions. Fig. 4 visually illustrates the docking interactions between EtfD and diacerein, levonadifloxacin, and gatifloxacin.
5. MD Simulation Analysis: MD is a computational technique modeling the dynamic behavior of molecular systems over time, treating all elements as flexible, and is typically used to further investigate the highest-ranked complex for detailed exploration [47]. Two widely used metrics for assessing structural fluctuations of macromolecules are Root-Mean-Square Deviation (RMSD) and Root-Mean-Square Fluctuations (RMSF). RMSD quantifies the average displacement of atoms from a reference structure, aiding in the analysis of time-dependent structural motions, often indicating stability or divergence, which may suggest simulation non-equilibration [41]. Whereas RMSF measures the displacement of specific atoms or groups from the reference structure, indicating protein flexibility during simulation [41]. RMSD and RMSF gauge protein mobility via rigid body alignment to a reference, but their sensitivity to fluctuating subsets can inflate values, potentially misrepresenting overall dynamics (Fig. 5a, b) [41]. The Radius of Gyration (RoG) is pivotal in assessing amino acid residue packing for protein stability, with conformational changes upon ligand binding making its calculation crucial for predicting macromolecular structural activity and stability (Fig. 5c) [42,43]. Hydrogen bonds and hydrophobic interactions are critical for stabilizing macromolecules, influencing binding affinity, drug efficacy, and providing essential insights for drug design and behavior prediction, with their paramount role in stabilizing the protein-ligand complex significantly impacting drug-target binding (Fig. 5d) [43].
a) Diacerein: During the first 20 ns, the RMSD of the backbone of EtfD complexed with diacerein exhibited a gradual increase, stabilizing thereafter with minor fluctuations within acceptable limits. The average RMSD of the backbone of EtfD, excluding IDRs, was calculated as 2.041 ± 0.459. Conversely, the full protein backbone RMSD displayed significant fluctuations, largely attributed to the IDRs, with a mean value of 5.803 ± 3.731. Comparison of the two RMSD values indicates overall stability of the protein structure throughout the simulation, except for the IDRs which exhibited the highest fluctuations. Additionally, RMSF analysis confirmed this observation, with residues excluding IDRs displaying RMSF values of 2.014 ± 0.601, while IDRs exhibited considerably higher RMSF values (11.511 ± 4.136). The Radius of Gyration (RoG) value for the bound complex was 34.094 ± 0.396, suggesting stability throughout the simulation, with minor fluctuations observed between 30 ns and 45 ns that returned to the averaged value in the final 5 ns. The presence of numerous hydrogen bonds between diacerein and EtfD throughout the simulation, involving residues HID74, ARG133, ARG146, THR296, and ARG300, underscores the high stability of the complex, with sustained interactions observed for ARG133 and ARG300.
b) Levonadifloxacin: The RMSD of the EtfD backbone, excluding IDRs, showed a gradual increase over the initial 20 ns, followed by stability until 30 ns, with subsequent minor fluctuations until 45 ns, returning to levels observed between 20 and 30 ns towards the end of the simulation. The average RMSD of the EtfD backbone, excluding IDRs, was 1.913 ± 0.328 Å. Conversely, RMSD of the full protein's backbone exhibited notable fluctuations primarily due to IDRs, with a recorded value of 5.219 ± 3.680 Å, highlighting the overall stability of the protein structure throughout the simulation, except for IDRs with high fluctuations. The RMSF analysis further validated these observations, showing residues excluding IDRs with an RMSF value of 1.935 ± 0.685 Å, while residues within IDRs displayed notably higher RMSF values of 10.166 ± 4.133 Å. Additionally, the Radius of Gyration (RoG) value for the bound complex was determined to be 33.675 ± 0.270 Å, indicating stability throughout the simulation duration. LEU64, TRP67, PRO70, ARG133, ARG146, HIE233, and ARG300 amino acids formed hydrogen bonds with levonadifloxacin during the simulation. Notably, HIE233, ARG146, PRO70, ARG133, and ARG300 maintained consistent interactions with levonadifloxacin throughout, with HIE233 being the most prominent, followed by ARG146, PRO70, ARG133, and ARG300.
c) Gatifloxacin: The RMSD of the EtfD backbone, excluding IDRs, increased gradually in the initial 10 ns, stabilized until 45 ns with minor fluctuations, followed by a notable increase. The average RMSD of the EtfD backbone with gatifloxacin was 2.067 ± 0.424 Å, indicating stability, while the full protein backbone exhibited high fluctuations primarily due to IDRs (4.127 ± 3.183 Å), stabilizing from 25 ns onwards with a temporary decrease at 35 ns. Comparative analysis showed overall protein structure stability except for IDRs, supported by RMSF values (residues excluding IDRs: 1.850 ± 0.521 Å, IDRs: 10.532 ± 4.037 Å). RoG value (33.533 ± 0.393 Å) indicated stability in the bound complex throughout the simulation. Notably, HID74, ARG133, ARG146, HID151, ASN152, ALA155, TRP156, HID233, and ARG300 residues formed significant hydrogen bonds with gatifloxacin, highlighting the high stability of this complex. Particularly, ARG300 exhibited multiple hydrogen bonds with the drug, followed by HID74 and ARG146. These
6. Binding free energy estimation: To validate the drugs' affinity to EtfD, MM-GB/PBSA post-simulation processing was conducted to obtain different free energies of the complexes (Table 2).
7. ADMET analysis: Table 3 provides a comprehensive overview of the pharmacokinetics of the screened compounds, encompassing druglikeness, medicinal chemistry, and various toxicity analyses. The ADMET analysis of diacerein unveiled quinone_A (PAINS) and phenol_ester (Brenk) alerts, suggesting assay interference and chemical instability, while levonadifloxacin and gatifloxacin exhibited no alerts or issues While effective at identifying many aggregators and assay artifacts, the PAINS rule's broad application, absence of mechanism exploration, and incomplete validation contribute to its low precision and limited scope for detection [48]. Furthermore, the presence of a Brenk violation in diacerein due to the phenol ester group is mitigated by its metabolic conversion to rhein, which lacks such problematic substructures, thus these alerts can be safely ignored.
8. Anti-TB sensitivity prediction: Certainly: Traditional drug testing against Mtb is challenging due to slow bacilli evolution, but machine learning tools like mycoCSM predict compound sensitivity preemptively (Table 4). The predicted minimum inhibitory concentration (MIC) values for levonadifloxacin and gatifloxacin were lower than for isoniazid but higher than for rifampicin, whereas diacerein had a lower MIC than both. Levonadifloxacin and gatifloxacin showed better predicted penetration of necrotic foci than rifampicin, while diacerein was comparable to rifampicin. This indicates their potential efficacy against Mtb, underscoring their suitability as anti-TB compounds.