Sage Journals: Discover world-class research

Abstract

Over the last decade, Mycobacterium tuberculosis has mutated into a putative ‘superbug’, as treatments against it have failed due to increasing antimicrobial resistance. As a result, the rising incidence of multidrug-resistant tuberculosis (MDR-TB) is posing a significant public health threat, thus, the need to develop effective drugs for MDR-TB has become an urgent priority. To identify new drug candidates for the treatment of MDR-TB, the present study was based on mycobacterial shikimate kinase (MtSK) as the pharmacological target. One hundred potential MtSK inhibitors were identified from literature and database searches to identify compounds that were designed to specifically function as MtSK antagonists. The ADME properties of these compounds were evaluated by using the SwissADME web tool. ProTox-II software was also used to investigate any potential endocrine disrupting effects, mediated through their interaction with oestrogenic and/or androgenic receptors. This study also aimed to predict LD₅₀ values of potential drug candidates that would be active against the standard H37Rv strain of M. tuberculosis, by using the ProTox-II in silico tool. The molecules for which no structural hazard alerts were identified with these software tools were further subjected to molecular docking analyses and molecular dynamic simulations to estimate their ability to interact with the MtSK enzyme. Preliminary results from SwissADME indicated that 30 molecules were drug-like, due to their physicochemical and pharmacokinetic properties. However, subsequent analysis with ToxTree and ProTox-II indicated that only three of these 30 drug-like molecules were suitable for taking forward into further in vitro experiments. This study, which is based on the use of commonly used open-source in silico tools, identified new MtSK ligands for potential use in the development of new drugs for the therapeutic management of tuberculosis. An initial prediction of their safety profile was also generated.

Keywords

endocrine disruption molecular docking molecular dynamic simulation shikimate kinase toxicoinformatics

Introduction

Tuberculosis (TB) is the second most common infectious disease in the world, behind COVID-19 (and above HIV/AIDS); it is the 13th largest cause of mortality worldwide.¹ As per a recent World Health Organisation (WHO) report, an estimated 10 million people fell ill with TB worldwide in 2020, of which 1.1 million were children, and the mortality was estimated to be 1.5 million.¹ Over the last decade, Mycobacterium tuberculosis has mutated into a putative ‘superbug’, as treatment methods and regimens have failed in the face of increasing antimicrobial resistance (AMR).² Multidrug-resistant tuberculosis (MDR-TB) continues to be a public health emergency, and patients with MDR-TB have limited treatment options for good recovery. Thus, the need to develop drugs for MDR-TB and find new pharmacological targets has become urgent in recent years.

Shikimate kinase (MtSK), an enzyme that belongs to the nucleoside monophosphate (NMP) kinase family, has been identified as an important target in the mycobacterial proteome, due to its novel presence within the tubercle pathogen and its absence in mammalian cells.³ It plays a significant role in the development and maintenance of the atypical mycobacterial cell wall, which is one of the preliminary defence systems that M. tuberculosis employs against therapeutic drugs.⁴ In M. tuberculosis, MtSK is the fifth enzyme in the shikimate pathway cascade, which catalyses a phosphate transfer from ATP to the shikimate anion to produce shikimate-3-phosphate (S3P) and ADP.⁵ It is responsible for the synthesis of the aromatic amino acids phenylalanine, tyrosine and tryptophan, which play an important role in the maintenance of the integrity and structure of the mycobacterial cell wall. Strategically designing molecules to inhibit the bioactivity of MtSK is an approach currently under investigation for the treatment of TB.⁶ A specific set of structural scaffolds could be identified as being probable pharmacophores, and then molecular docking analyses could be carried out to assess any potential antagonistic effects of these ligands on MtSK.

Several tools are available to evaluate the Absorption, Distribution, Metabolism and Elimination (ADME) characteristics of thousands of drug candidates, to select the few that show promise for further development through the drug discovery process. The SwissADME tool provides key characteristics of compounds, such as molecular weight, number of rotatable bonds, hydrogen bond acceptor and donor potential, and total polar surface area.⁷ It also predicts lipophilicity, water solubility and pharmacokinetic properties, such as gastrointestinal absorption, blood–brain barrier (BBB) permeation and interaction with the P-glycoprotein (P-gp) efflux pump. SwissADME can screen large chemical libraries by using multiple filters, including Lipinski, Ghose, Veber, Muegge and Egan, to evaluate lead-likeness and determine their druggability (the likelihood of a drug-like compound to modulate or inhibit an interaction between two proteins).⁸

ToxTree is an in silico tool that can identify potential hazards, using information based on molecular structure.⁹ Proposed lead molecules can be toxicologically profiled by using ToxTree, which uses a decision-tree approach to examine the potential toxicological implications of the structural moieties in the query compounds. The lead molecules are evaluated against alerts for properties such as mutagenicity, genotoxic and non-genotoxic carcinogenicity, micronucleation, chromosomal aberrations, and the ability to bind DNA and protein.

Molecular docking analyses present still images of complexed protein–ligand co-crystals, whereas molecular dynamic simulations display physical movements and chemical interactions within a protein–ligand co-crystal over a fixed time-period. Molecular docking studies prove useful in predicting the extent of the interaction and complexation between different biomolecules, and to estimate the stability of these complexes. In contrast, molecular dynamic simulations provide precise information on the chemical structures, conformational changes, and dynamics of biological molecules and their complexes. Therefore, the combination of molecular docking analysis and molecular dynamic simulation permits quick screening of large libraries, and the simultaneous calculation of interaction energies to accelerate drug discovery processes.^10,11

In the current study, a literature search in PubMed was performed, in order to identify compounds that were designed to specifically function as MtSK inhibitors. One hundred molecules were randomly selected from the 767 compounds obtained from the literature search. These compounds were known molecular scaffolds — such as benzothiazole, benzimidazole, imidazole, triazole, phenylpiperidine, morpholino-piperazine, oxazole, benzoxazole, naphthalene-sulphonic acid and xanthene. The molecules were then evaluated for their pharmacokinetic and drug-like properties by using the SwissADME tool.

The thirty most promising molecules were then subjected to toxicological profiling with ToxTree software. The MtSK inhibitors that had no structural alerts identified in ToxTree were then screened for any potential endocrine-disrupting effects (mediated via their potential interaction with oestrogenic and/or androgenic receptors), by using ProTox-II.

An attempt to predict LD₅₀ values of potential anti-TB drug candidates was also made with ProTox-II, which employs a random forest algorithm for predicting the toxicities of small molecules, by annotating structures that may be potentially hazardous.

The 30 drug-like molecules were also subjected to molecular docking analyses, in order to predict the extent of their interaction and complexation with the co-crystallised structure of the MtSK enzyme. For the lead-like molecules that were indicated to be potent antagonists of the MtSK enzyme, molecular dynamic simulations were then undertaken to estimate the stabilities of the molecule–target complexes.

Methods

Identification of possible lead molecules

A literature survey was performed to identify compounds that were designed to specifically function as MtSK inhibitors by using the terms “MtSK inhibitors”, “Mycobacterium tuberculosis shikimate kinase inhibitors” in PubMed; their ZINC IDs¹³ and PubChem IDs¹⁴ were retrieved. The literature search identified 767 compounds, of which 100 compounds were randomly selected. The structures were downloaded in 2-D format from the respective websites, and their SMILES (Simplified Molecular Input Line Entry System) were generated by using the online tool SMILES translator (https://cactus.nci.nih.gov/translate/). SMILES is an easy-to-use chemical notation that allows the user to input a chemical structure in a way that can be processed by software.¹⁵

Prediction of pharmacokinetic properties and drug-likeness by using SwissADME

The Swiss Institute for Bioinformatics’ online tool, SwisADME (http://www.swissadme.ch/), was used to predict the PK and drug-likeness properties of the molecules and to analyse their druggability and lead-likeness by creating a bioavailability radar for the compounds. The SMILES were entered into the tool’s input box on the web page, and the tool generated the results independently.

The prediction of toxicological potential and threat aversion by using ToxTree

The toxicity of a molecule/ligand can be determined by its structural characteristics, and the ToxTree software forecasts specific toxicity potential according to the different structural moieties within the molecule. Structure–activity relationships, incorporated within the software, were used to identify possible toxicological outcomes. The following decision trees were chosen:

Kroes TTC (Threshold for Toxicological Concern)

The prediction of the threshold for toxicological concern of a molecule provides a pragmatic approach to assess the potential of a molecule/ligand to manifest toxicity in a physiological system above a certain amount of regulated intake.¹⁶

Carcinogenicity assessment (genotoxicity, non-genotoxicity, mutagenicity)

The potential of a molecule to induce cancer in a physiological system is a functional attribute of its structural components. The decision tree identifies the structural moieties within a molecule that could be a possible concern with regard to carcinogenicity or mutagenicity within a physiological system, indicating the possibility that such moieties could induce toxicities via genotoxic or non-genotoxic pathways.^17,18

In vitro mutagenicity assay (Ames Assay) alerts, as per the Istituto Superiore di Sanità (ISS) rule base

This rule base provides structural alerts based on the presence of structural moieties that may be positive for inducing mutations in Salmonella typhimurium.¹⁹

Structural alerts for the micronucleus assay in rodents

The ability of a ligand to induce micronuclei is considered to be a manifestation of toxicity, as it indicates the possibility of chromosomal aberrations. This rule base identifies structural alert moieties present in ligands that may possess the potential to induce micronucleus formation.²⁰

DNA binding alerts

The structural attributes of the lead molecules were assessed by using the DNA binding rule base, in order to identify whether the ligands have the potential to form covalent bonds with DNA within a physiological system. Such covalent interactions between a ligand and DNA are irreversible in nature, hampering the development of an effective therapeutic agent.²¹

Protein binding alerts

The structural attributes of the lead molecules were also assessed by using the protein binding rule base, in order to identify the potential to form covalent bonds with proteins within a physiological system — such covalent interactions would be unfavourable. The SMILES generated were inputted into the software, and the results were generated by manually selecting the decision tree to be applied.²²

Prediction of endocrine disruption potential

The 30 molecules deemed most promising following the initial analyses were further screened in silico for their potential to interact with oestrogenic and/or androgenic receptors and cause functional endocrine disruptions within a physiological system. The online tool, ProTox-II (https://tox-new.charite.de/protox_II/), was employed for this purpose. SMILES were inputted into the input box, and the receptors were chosen sequentially to estimate the bioactive potential of the lead molecules and to interpret their impact on the endocrine proteome. The targeted endocrine receptors included: the androgen receptor (AR), androgen receptor-ligand binding domain (AR-LBD), oestrogen receptor-α (ER-α), oestrogen receptor-ligand binding domain (ER-LBD) and peroxisome proliferator-activated receptor-γ (PPAR-γ).^12,23

Prediction of LD₅₀ values

ProTox-II is a virtual in silico tool that estimates median lethal doses (LD₅₀ values) and categorises compounds into six classes ranging from Class I (LD₅₀ ≥ 5 mg/kg, i.e. fatal if swallowed), to Class VI (LD₅₀ ≤ 5000 mg/kg, i.e. non-toxic); each of the classes is indicative of a compound’s toxicological profile (Table 1). SMILES for the compounds were inserted into the input widget and the results were generated.^12,23

Table 1.

Toxicological classification for LD₅₀ values.

Class	Inference
Class I	Fatal if swallowed (LD₅₀ ≤ 5 mg/kg)
Class II	Fatal if swallowed (5 < LD₅₀ ≤ 50 mg/kg)
Class III	Toxic if swallowed (50 < LD₅₀ ≤ 300 mg/kg)
Class IV	Harmful if swallowed (300 < LD₅₀ ≤ 2000 mg/kg)
Class V	May be harmful if swallowed (2000 < LD₅₀ ≤ 5000 mg/kg)
Class VI	Non-toxic (LD₅₀ > 5000 mg/kg)

Molecular docking analyses and molecular dynamic simulations

Selection and processing of the macromolecule

The 3-D structure of the M. tuberculosis MtSK target protein, complexed with shikimic acid, was procured from the public database, RCSB PDB (https://www.rcsb.org/). After a careful evaluation of the selection parameters (i.e. resolution of the protein structure and Ramachandran outliers), 2IYQ (PDB DOI: 10.2210/pdb2IYQ/pdb) was selected as the appropriate macromolecule. It had a crystallographic resolution of 1.8 Å and was complexed with shikimic acid and ADP.²⁴ The macromolecule was then processed by using the Discovery Studio tool. The addition of H atoms was carried out and was followed by the equitable distribution of respective charges of the amino acid groups across the structure of the macromolecule. Energy minimisation of the macromolecule was then carried out with Chimera software.²⁴ The macromolecule was then submitted to CASTp (http://sts.bioe.uic.edu/castp/index.html?1bxw) for the prediction of the active site of the enzyme, and the final processed file of the macromolecule 2IYQ was used for the molecular docking analyses.

Retrieval of ligands

Public libraries such as the ZINC database (https://zinc.docking.org/) and the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) were used to procure the structures of the 30 ligands (out of the initial 100 ligands) that were considered the most promising following the preliminary pharmacokinetic and toxicological screening. The 3-D structures of the 30 ligands were downloaded in ‘.sdf’ format and the ligand structures were optimised by using the energy minimisation module on the open-source molecular docking analysis tool, PyRx. Further, the validation of the structures of the 30 ligands was carried out by stereochemical modifications, such as the addition of valence hydrogen, and optimisation of the bond lengths and bond angles. The ligands were subsequently converted to ‘.pdbqt’ format for use in the molecular docking analyses.

Molecular docking analyses

Molecular docking analyses were carried out by using the AutoDock Vina Module 1.1.2. implemented on the PyRx software.²⁵ The macromolecule 2IYQ was similarly uploaded to the AutoDock Vina, designated as the macromolecule, and converted to the ‘.pdbqt’ format. Docking parameters, such as the algorithm for docking and grid space for docking ligands onto the macromolecule, were finalised and the docking analyses were carried out for the 30 ligands that could potentially act as MtSK antagonists. Molecular docking analyses were carried out by using the Lamarckian Genetic Algorithm integrated within the AutoDock Vina module.^25,26 Analyses for the visualisation of ligand–protein interactions, both in 2-D and 3-D, were carried out with the Discovery Studio software (see online Supplemental Material).

Molecular dynamic simulations

After performing a virtual screening for suitable binding affinity in PyRx, and for potential toxicity in ToxTree and ProTox-II, three promising lead molecules were identified — namely, ZINC15707201 (M1), ZINC11790367 (M2) and ZINC588497 (M23). These ligands were tested for co-crystallisation stability by performing a 10-nanosecond simulation and then a high-throughput 100-nanosecond simulation with the latest version of GROMACS software (2023 series), which gives fast and reliable results for biomolecules (see online Supplemental Material). These simulations were subsequently analysed by using xmgrace and UCSF Chimera software to investigate whether these ligands would diffuse from the MtSK target protein over time. By visualising the whole trajectory of the interaction in UCSF Chimera, it was possible to confirm that, theoretically, these molecules were able to substantially interact with the MtSK target protein.^6,10,11

Results and discussion

Prediction of pharmacokinetic properties and drug-likeness by using SwissADME

To identify the molecules most likely to be safe and druggable, each of the initial 100 molecules was subjected to an evaluation of its pharmacokinetic and drug-like properties, including the bioavailability profile, Log P value, ability to breach the BBB, GI absorption and ability of P-gp to eliminate the substance from the cranial environment. Of these 100 molecules, 30 were found to have physicochemical properties falling within the pink area of the bioavailability radar in Figure 1 — i.e. within a suitable physicochemical space for oral bioavailability. The pink area represents the optimal range for each property, namely:

— lipophilicity (LIPO): a XLOGP3 value of between −0.7 and +5.0;

— size (SIZE): a molecular weight (MW) of between 150 and 500 g/mol;

— polarity (POLAR): a total polar surface area (TPSA) of between 20 and 130 Å²;

— solubility (INSOLU): a log S value not higher than 6;

— saturation (INSATU): the fraction of sp³ hybridised carbons not less than 0.25; and

— flexibility (FLEX): no more than 9 rotatable bonds.⁷

Figure 1.

Selection of drug-like molecules after the pharmacokinetic analysis.

The data ranges for each of the analysed properties, and the proportion of the initial 100 molecules falling within each range, are shown in Table 2a.

Table 2.

Analysis of the 100 potential MtSK inhibitors, by using SwissADME.

a) Physico-chemical properties
Property	Range	Proportion of molecules
Number of H-bond donors	< 3	76%
	3	12%
	> 3	12%
Number of H-bond acceptors	1–4	37%
	5–8	57%
	9–11	6%
Molecular weight (g/mol)	150–300	16%
	301–450	53%
	451–600	31%
Total polar surface area (Å²)	0–50	3%
	51–100	50%
	101–150	38%
	151–200	5%
	201–250	4%
Log P (O/W)	< 2.5	27%
	2.5	1%
	> 2.5	72%

b) ADME properties
Property	Prediction	Proportion of molecules
Gastrointestinal absorption	High absorption	73%
Gastrointestinal absorption	Low absorption	27%
Blood–brain barrier permeation	Permeant	81%
Blood–brain barrier permeation	Impermeant	19%
P-glycoprotein substrate	Substrate	59%
P-glycoprotein substrate	Non-substrate	41%

c) Drug-likeness
Screening rules	Prediction	Proportion of molecules
Lipinski’s ‘rule of five’ for drug-likeness	Drug-like	95%
Lipinski’s ‘rule of five’ for drug-likeness	Not drug-like	5%
Veber filter for drug-likeness	Drug-like	79%
Veber filter for drug-likeness	Not drug-like	21%
Egan filter for drug-likeness	Drug-like	68%
Egan filter for drug-likeness	Not drug-like	32%
Ghose filter for drug-likeness	Drug-like	60%
Ghose filter for drug-likeness	Not drug-like	40%
Muegge filter for drug-likeness	Drug-like	72%
Muegge filter for drug-likeness	Not drug-like	28%

The hydrogen bond acceptor and donor counts within a compound are vital physicochemical properties that are used to estimate its oral bioavailability. It is commonly believed that these characteristics influence passive diffusion across biological membranes during absorption and distribution. Lipinski’s ‘rule of five’ and the Muegge filter recommend that, for a compound to exhibit drug-like properties, it should have no more than ten H-bond acceptors and no more than five H-bond donors. As Table 2a shows, of the initial 100 molecules, 12% had more than three H-bond donor moieties and the majority of the compounds (57%) possessed between five and eight H-bond acceptors.

The permeability of a compound, and thus its bioavailability, is related to its molecular weight. High molecular weight compounds tend to have poor bioavailability, due to their limited permeability across membranes. It is notable that the molecular weight of every analysed compound was less than 600 Da, and was skewed toward 301–450 Da — 53% of the compounds fell within this range. Only 16% of the compounds studied had a low molecular weight (i.e. between 150–300 Da), whereas the remaining 31% were between 451–600 Da.

The Total Polar Surface Area (TPSA) is a measure of a compound’s surface that is represented by polar atoms. It correlates with passive diffusion, and thus the permeability and bioavailability of a particular compound. As mentioned above, the optimal range for the TPSA, according to the SwissADME, is between 20 and 130 Å². TPSA prediction data suggested that 50% of the compounds analysed in the current study had a TPSA of between 51–100 Å², and 38% of them had TPSA values within the 101–150 Å² range.

The lipophilicity of a compound is determined by its partition coefficient and is expressed as a Log P value. This is the logarithm of the ratio of solubilities of a particular compound in water and octanol. The higher the Log P value, the higher the lipophilicity of a compound, and the desirable range (as per SwissADME) is between −0.7 and +5.0. The Log P value of a compound indicates its absorption, transportation and distribution potential, and thus its overall bioavailability. A low water–octanol lipophilicity coefficient (i.e. 1.5–2.5) was predicted for 27% of the molecules analysed, and a value above 2.5 was predicted for 72% of the compounds.

The predicted ADME properties (Table 2b) showed that 73 of the 100 molecules analysed would have high GI absorption, and that 81% would be able to penetrate the BBB; 59% were predicted to be potential P-gp substrates. Understanding whether or not compounds are substrates of P-gp is crucial for evaluating their active efflux across biological membranes. This is especially important for processes such as transport from the gastrointestinal wall to the lumen, or across the BBB. A primary function of P-gp is to safeguard the central nervous system (CNS) from foreign substances.

Lipinski’s ‘rule of five’ showed that 95% of the analysed compounds were drug-like. The Veber filter indicated that 79% of the compounds had drug-like structural attributes, while the Egan filter showed 68% of compounds to be drug-like (Table 2c). The Ghose and Muegge filters indicated that 60% and 72% of the molecules, respectively, were drug-like.

After analysis of the 100 molecules, the bioavailability radar on the SwissADME tool (Figure 1) was used to systematically filter and select 30 molecules with the most promising drug-like properties. These 30 selected molecules (see Table 3) were further assessed for toxicity and endocrine disruption potential in ToxTree and ProTox-II, respectively.

Table 3.

The top 30 drug-like compounds, identified from their pharmacokinetic and drug-like properties and their structural identifiers.

Molecule No.	Structural identifier (Molecule ID)	Molecular structure
M1	ZINC15707201
M2	ZINC11790367
M3	ZINC8442094
M4	ZINC633992
M5	ZINC16192643
M6	ZINC11881196
M7	ZINC393534
M8	ZINC34797195
M9	CHEMBL1894076
M10	ZINC20453061
M11	ZINC17982744
M12	ZINC1576968
M13	ZINC15675581
M14	ZINC20464408
M15	ZINC737165696
M16	ZINC19366016
M17	ZINC197090
M18	ZINC5489375
M19	ZINC19203136
M20	ZINC13545928
M21	ZINC167795816
M22	CHEMBL4060253
M23	ZINC5884947
M24	ZINC204453061
M25	ZINC44036990
M26	ZINC20720307
M27	ZINC20720315
M28	ZINC5605047
M29	ZINC71963683
M30	ZINC45977224

Prediction of toxicological potential and threat aversion by using ToxTREE

The 30 selected drug-like molecules were subjected to further analyses by using decision trees such as Kroes TTC, carcinogenicity assessment, in vitro mutagenicity (Ames Assay), in vitro micronucleus assay, and DNA binding and protein binding alerts (Table 4). Of the 30 lead molecules, 28 triggered at least one structural alert for genotoxic carcinogenicity from the Kroes TTC analysis, while only three were concluded to be potentially genotoxic from the carcinogenicity decision tree. None of the 30 drug-like molecules were reported to have mutagenic potential; three molecules had two alerts, 25 had one alert, and two had no alerts for chromosomal aberrations, as highlighted in the relevant decision tree results in Table 4.

Table 4.

Summary of the ToxTree analysis for the top 30 drug-like compounds.

Molecule No.	Decision trees for toxicological potential and threat aversion
Molecule No.	Carcinogenicity^a	Kroes TTC	Ames Assay^b	In Vitro Micronucleus Assay	DNA binding	Protein binding
M1	1 SA for non-genotoxic carcinogenicity	N	N	1 SA for chromosomal aberration	MA: Covalent binding	MA, SN2 aliphatic substitution
M2	N	N	N	N	MA: Covalent binding	MA: Covalent binding
M3	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M4	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M5	N	1 SA for genotoxic carcinogenicity	N	2 SAs for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M6	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M7	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA, SN2 aliphatic substitution
M8	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA, SN2 aliphatic substitution
M9	N	1 SA for genotoxic carcinogenicity	N	2 SAs for chromosomal aberration	MA, SN2 aliphatic substitution	MA: Covalent binding
M10	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA, SN2 aliphatic substitution
M11	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	N	N
M12	N	1 SA for genotoxic carcinogenicity	N	N	MA, SN2 aliphatic substitution	MA: Covalent binding
M13	N	2 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA, SN2 nucleophilic aliphatic substitution	MA: Covalent binding
M14	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA, SN2 nucleophilic aliphatic substitution	MA, SN2 aliphatic substitution
M15	2 SAs for genotoxic carcinogenicity	2 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	N	N
M16	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	N
M17	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	N	N
M18	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M19	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M20	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	SN1 nucleophilic addition	SN1 nucleophilic addition
M21	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M22	N	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	Nucleophilic aromatic substitution, MA
M23	N	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	N	N
M24	N	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	Nucleophilic aromatic substitution, nucleophilic aliphatic substitution
M25	1 SA for genotoxic carcinogenicity	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M26	N	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M27	N	2 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M28	N	1 SA for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	N	N
M29	N	1 SAs for genotoxic carcinogenicity	N	1 SA for chromosomal aberration	MA: Covalent binding	MA: Covalent binding
M30	2 SAs for genotoxic carcinogenicity	1 SA for genotoxic carcinogenicity	N	2 SAs for chromosomal aberration	SN1 nucleophilic addition	SN1 nucleophilic addition

^a‘Carcinogenicity’ includes genotoxicity and non-genotoxicity.

^bIn Vitro Ames Mutagenicity Assay.

MA = Michael Acceptor; N = No Alert; SA = Structural Alert; TTC = Threshold for Toxicological Concern. Bold entries indicate toxicological potential.

The highest probability of binding to DNA and interfering with its structure and function applied to six of the 30 drug-like molecules, while 19 molecules showed lower potential for interacting with DNA; five compounds were predicted to be devoid of structural moieties likely to interact with DNA and hamper physiological function. The protein binding decision tree indicated that nine molecules had two structural alerts for interaction with skin proteins, while 15 molecules elicited one structural alert for protein interaction. The 24 molecules that triggered at least one structural alert were all Michael acceptors and were capable of binding covalently to skin proteins. In contrast, six molecules were predicted to be free of structural moieties involved in the formation of such covalent bonds.

Prediction of endocrine disruption potential and hepatotoxicity

After SwissADME profiling, 30 molecules were deemed to be drug-like. These molecules were subjected to further assessment, in order to predict their possible impact on the endocrine system. This assessment, using the online tool ProTox-II (https://tox-new.charite.de/protox_II/), was based on the molecule’s ability to bind to oestrogenic and/or androgenic receptors (see Table 5). Five endocrine receptors were evaluated in this study: the androgen receptor (AR), androgen receptor-ligand binding domain (AR-LBD), oestrogen receptor-α (ER-α), oestrogen receptor-ligand binding domain (ER-LBD), and peroxisome proliferator-activated receptor-γ (PPAR-γ). Each molecule was predicted to be either ‘Active’ or ‘Inactive’, in relation to interaction with the target receptors. The molecules were classified as ‘endocrine disrupting’ or ‘endocrine non-disrupting’, according to the total number of receptors with which the molecule showed significant interaction (i.e. which would be classified as ‘Active’). No molecule was identified as a potential endocrine disruptor, according to these criteria. However, M21 was classified as ‘Inactive’ with regard to its interaction with the oestrogen receptor-α, but with a low probability of incidence (0.68) — hence its predicted non-interaction with the oestrogen receptor-α (i.e. ‘Inactive’) should be viewed with caution; this compound is highlighted in bold in Table 5.

Table 5.

The ProTox-II analysis for the prediction of hepatotoxicity and endocrine disruption potential.

Molecule No.	Hepatotoxicity^a	Interaction with oestrogen receptor-α^b
M1	Inactive (0.59)	Inactive (0.87)
M2	Inactive (0.76)	Inactive (0.89)
M3	Inactive (0.59)	Inactive (0.85)
M4	Inactive (0.57)	Inactive (0.88)
M5	Inactive (0.53)	Inactive (0.89)
M6	Inactive (0.79)	Inactive (0.89)
M7	Inactive (0.84)	Inactive (0.75)
M8	Inactive (0.51)	Inactive (0.86)
M9	Active (0.54)	Inactive (0.93)
M10	Inactive (0.57)	Inactive (0.84)
M11	Inactive (0.75)	Inactive (0.88)
M12	Inactive (0.65)	Inactive (0.87)
M13	Inactive (0.66)	Inactive (0.90)
M14	Inactive (0.64)	Inactive (0.91)
M15	Inactive (0.57)	Inactive (0.87)
M16	Active (0.57)	Inactive (0.89)
M17	Inactive (0.62)	Inactive (0.86)
M18	Inactive (0.53)	Inactive (0.94)
M19	Inactive (0.77)	Inactive (0.93)
M20	Inactive (0.91)	Inactive (0.94)
M21	Inactive (0.75)	Inactive (0.68)
M22	Inactive (0.68)	Inactive (0.90)
M23	Inactive (0.52)	Inactive (0.87)
M24	Inactive (0.61)	Inactive (0.88)
M25	Inactive (0.90)	Inactive (0.89)
M26	Inactive (0.78)	Inactive (0.82)
M27	Inactive (0.85)	Inactive (0.89)
M28	Inactive (0.84)	Inactive (0.91)
M29	Inactive (0.68)	Inactive (0.83)
M30	Inactive (0.56)	Inactive (0.97)

^aThe values in parentheses in this column indicate the probability of incidence of the classified effect — either ‘Inactive’ in terms of hepatotoxicity, or ‘Active’. A value of < 0.7 represents a ‘low’ probability of incidence; a value of > 0.7 represents a ‘high’ probability. The bold entries in this column indicate a potential for hepatotoxic activity (i.e. drug-induced liver injury; DILI).

^bThe values in parentheses in this column indicate the probability of incidence of the classified effect — i.e. ‘Inactive’ with regard to their interaction with the oestrogen receptor-α. A value of < 0.7 represents a ‘low’ probability of incidence; a value of > 0.7 represents a ‘high’ probability of incidence. Hence, the bold entry indicates that this ‘Inactive’ classification has a low probability of incidence (< 0.7).

Interactions with the androgenic receptor, androgenic receptor-ligand binding domain, oestrogenic receptor-ligand binding domain, and PPAR-γ were also analysed and deemed to be insignificant — i.e. all were classified as ‘Inactive’ with regard to their interactions, with (high) probabilities of incidence of between 0.75 and 1.0; thus these data are not shown in the Table.

The same tool, ProTox-II, was also used to predict the potential for hepatotoxicity (i.e. drug-induced liver injury; DILI). Two molecules (M9 and M16) were highlighted by ProTox-II as having the potential to cause DILI, albeit at a low probability of incidence (< 0.7); these are indicated in bold in Table 5. The remaining 28 molecules were classified as ‘Inactive’ with regard to their potential hepatotoxic activity — 17 with a high probability of incidence (> 0.7) and 11 with a low probability (< 0.7).

Prediction of LD₅₀

Computational studies can rely on predefined test sets to annotate the training set of molecules and predict the LD₅₀ values, based on the structural components of the molecules. This approach can lead to the more efficient selection of lead molecules, prior to in vitro analyses. In the evaluation carried out with ProTox-II in the current study (see Table 6), out of the 30 selected molecules, none were placed in Class I or Class II, three were classified as Class III, 19 molecules were identified as Class IV, five molecules were placed into Class V, and three were considered, potentially, to be non-toxic at doses higher than 5000 mg/kg (i.e. Class VI) — see Table 1 for the various toxicological classification dose ranges, Class I to VI.

Table 6.

Predicted LD₅₀ values of the 30 compounds deemed to be drug-like.

Molecule No.	Predicted LD₅₀ (mg/kg)	Predicted Class
M1	2000	IV
M2	300	III
M3	500	IV
M4	500	IV
M5	1000	IV
M6	357	IV
M7	5000	V
M8	4000	V
M9	722	IV
M10	1500	IV
M11	5000	V
M12	11000	VI
M13	1000	IV
M14	1000	IV
M15	300	III
M16	2158	IV
M17	1000	IV
M18	940	IV
M19	8000	VI
M20	3330	V
M21	650	IV
M22	1200	IV
M23	1000	IV
M24	251	III
M25	1600	IV
M26	5000	V
M27	1750	IV
M28	1500	IV
M29	640	IV
M30	7710	VI

Molecular docking analyses

The 30 drug-like molecules subjected to molecular docking analyses each showed a high probability of interacting with the target enzyme, MtSK (Table 7). The AutoDock Vina software that was used for the docking analyses, scores binding affinity of a compound based on its calculated binding energy. A higher negative binding energy value indicates better binding affinity; compounds with positive binding energies are regarded as non-binders.

Table 7.

The binding affinity scores for the 30 drug-like molecules, derived from molecular docking analysis on PyRx.

Molecule No.	Calculated binding energy (kcal/mol)
M1	–7.9
M2	–7.3
M3	3.8
M4	–7
M5	–6.8
M6	–6.8
M7	–6.7
M8	–6.6
M9	–6.5
M10	–6.5
M11	–6.4
M12	–6.3
M13	–5.1
M14	–5
M15	–5
M16	–1.9
M17	–1.7
M18	–1.5
M19	0.6
M20	1
M21	2.5
M22	3.4
M23	–7.1
M24	7.3
M25	10.2
M26	11.9
M27	16.6
M28	20
M29	28.7
M30	32.9

The assessment was performed against the target enzyme, MtSK (2IYQ protein structure).

The molecular docking scores of the 30 drug-like molecules were compared, and those with the top three scores in terms of the calculated binding energy (i.e. < –7 kcal/mol) were subsequently chosen for the molecular dynamic simulations — namely, M1, M2 and M23.

Molecular dynamic simulations

From the molecular dynamic simulation data, the complex between M1 and 2IYQ appeared to be stable (Figure 2a and b). However, the lead molecule showed some deviation from the original binding site to a possible allosteric site, as noted from the rise in root-mean-square deviation delta values (δ-RMSD values). The complex formed by M2 and 2IYQ (Figure 2c and d) also seemed significantly energetic, as the interactions between the lead molecule and the crystalline structure of the MtSK enzyme show significantly deviating δ-RMSD values.

Figure 2.

RMSD and δ-RMSD plots of the interactions between the three most promising lead molecules and the MtSK target enzyme (2IYQ prostein structure).

The complex between M23 and 2IYQ (Figure 2e and f) was reportedly more stable than that formed with M1 and M2, as evidenced by the lower drift in δ-RMSD values. Despite some drifts observed in the δ-RMSD values, the complexes formed remained stable enough to ensure that the crystalline structure of the enzyme would not denature, as the degree of gyration was maintained for all of the structures.

Thus, the present study finally proposed three molecules to be the most promising candidates in terms of predicted safety profile and potency of MtSK inhibition — namely, M1, M2 and M23 — which could potentially be taken forward into further in vitro experiments.

Synthesis scheme for two of the proposed ligands

A proof-of-concept attempt was made to predict a scheme of synthesis for M1 and M2 by using a retrograde synthon approach to aid in vitro studies in the future.

The retrosynthesis prediction of M1 was performed with IBM RXN for chemistry webserver. The proposed chemical pathway for the synthesis of M1, deduced from the retrosynthesis prediction, is depicted in Figure 3. Step 1 in Figure 3a includes three different reactions. The first reaction is protection of the N-1 and N-7 of spinacine using isobutyrlchloroformate in tetrahydrofuran (THF) with the aid of diisopropylethylamine (DIPEA) at 40°C to form the N-1,N-7 protected (N-carbamoyl) spinacine derivative (not shown). The second reaction is heterocyclisation of the resulting N-1,N-7 protected spinacine derivative by N-hydoxyisobuyramidine with DIPEA in THF at 120°C to N-1,N-7-protected 1,2,4-oxadiazolylspinacine derivative (not shown). The third reaction is N-1,N-7-deprotection (removal of carbamoyl group) from N-1,N-7-protected 1,2,4-oxadiazolylspinacine derivative using trifluoracetic acid (TFA) in dichloromethane (DCM) to yield 6-[3-(propan-2-yl)-1,2,4-oxadiazol-5-yl]spinacine.²⁷ Step 2, shown in Figure 3b, is the condensation of oxadiazolylspinacine derivative obtained from Step 1 (Figure 3a) with ethyl 4-isocyanatobenzoate in acetonitrile (ACN) by using microwaves (MW, 200W) at 70°C to give the desired compound, i.e. M1.²⁸

Figure 3.

Retrosynthesis pathway prediction for M1.

The retrosynthesis pathway prediction for M2 was carried out based on a literature survey of known chemical reactions, and the proposed synthetic scheme obtained from the literature review is shown in Figure 4. Step 1 (Figure 4a) is the condensation of 3-pheylpropanoic acid with 1-(4-aminophenyl)piperidin-4-ol by using N,N´-dicyclohexylcarbodiimide (DCC) and dimethylaminopyridine (DMAP) in dimethyl sulphoxide (DMSO) at 0°C, to give the phenylpropanamide derivative.²⁹ Step 2 (Figure 4b) is the N-alkylation of (1-(pyrimidin-2-yl)piperidin-3-yl)methanamine with the phenylpropanamide derivative obtained in Step 1 with the use of a manganese catalyst complex, MnBr(CO)₂[NH(CH₂CH₂P(CH(CH₃)₂)₂)₂], and potassium tert-butoxide (t-BuOK) in toluene to give the desired compound, i.e. M2.³⁰

Figure 4.

Retrosynthesis pathway prediction for M2.

Conclusions

Strategies that target MtSK as new anti-TB therapies show potential but remain under-explored. Toxicoinformatics is an important element in the drug development process, and the emergence of structure-based predictions and read-across studies, as well as the integration of artificial intelligence in such investigations, has considerably accelerated drug discovery and development. Moreover, these new approaches confer economic, time-saving and ethical benefits, as they can be used for the initial evaluation of compounds and thus ensure that conventional in vivo studies are carried out only if the initial in silico results are promising. The present study, which employed a range of commonly used open-source in silico tools to predict the safety and efficacy of potential MtSK inhibitors, has led to the identification of three new MtSK ligands as possible drugs for the therapeutic management of TB. More in-depth analyses of these potential MtSK inhibitors are warranted in future studies.

Supplemental Material

Supplemental Material - Toxicological Profiling of Potential Shikimate Kinase Inhibitors Against Mycobacterium tuberculosis

Supplemental Material for Toxicological Profiling of Potential Shikimate Kinase Inhibitors Against Mycobacterium tuberculosis by Ashish Jhangiani, Vandana Panda, Aanchal Sukheja, Sneha Thomas, Piyush Dusseja, Siddhartha Pandya and Anand Chintakrindi in Alternatives to Laboratory Animals

Supplemental Material

Footnotes

Acknowledgment

The authors thank Prof. Urmila Joshi for her valuable advice to help improve the manuscript.

Author contributions

AJ contributed to the conceptualisation of the idea, literature review, drafting of the manuscript, and generation and procurement of the raw data. VP contributed to the conceptualisation of the idea, inferencing of the raw data, and writing and revision of the manuscript. AC contributed to the prediction of synthesis schemes for the molecules. SP contributed toward the molecular dynamic simulations performed for the three molecules and molecular docking analyses performed for the 30 molecules. AS, ST and PD contributed toward the procurement of raw data and literature review. All authors reviewed and approved the final manuscript.

Declaration of conflicting interests

The author(s) declare no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

Supplemental material for this article is available online.

References

WHO. Tuberculosis , https://www.who.int/news-room/fact-sheets/detail/tuberculosis (2021, accessed 7 August 2023).

Allué-Guardia

García

Torrelles

. Evolution of drug-resistant Mycobacterium tuberculosis strains and their adaptation to the human lung environment. Front Microbiol 2021; 12: 612675.

Coracini

Azevedo

. Shikimate kinase, a protein target for drug design. Curr Med Chem 2014; 21: 592–604.

Gordon

Simithy

Goodwin

, et al. Selective Mycobacterium tuberculosis shikimate kinase inhibitors as potential antibacterials. Perspect Medicinal Chem 2015; 7: 9–20.

Nunes

JES

Duque

de Freitas

, et al. Mycobacterium tuberculosis shikimate pathway enzymes as targets for the rational design of anti-tuberculosis drugs. Molecules 2020; 25: 1259.

Vianna

de Azevedo

. Identification of new potential Mycobacterium tuberculosis shikimate kinase inhibitors through molecular docking simulations. J Mol Model 2012; 18: 755–764.

Daina

Michielin

Zoete

SwissADME: A free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules. Sci Rep 2017; 7: 1–3.

Bakchi

Ambati

Ekambarapu

, et al. An overview on applications of SwissADME web tool in the design and development of anticancer, antitubercular and antimicrobial agents: A medicinal chemist’s perspective. J Mol C Struct 2022; 1259: 132712.

Rim

. In silico prediction of toxicity and its applications for chemicals at work. Toxicol Environ Health Sci 2020; 12: 191–202.

10.

Hollingsworth

Dror

. Molecular dynamics simulation for all. Neuron 2018: 99: 1129–1143.

11.

Aghajani

Farnia

, et al. Molecular dynamic simulations and molecular docking as a potential way for designed new inhibitor drug without resistance. TANAFFOS (Respiration) 2022; 21: 1–4.

12.

Banerjee

Eckert

Schrey

, et al. ProTox-II: A webserver for the prediction of toxicity of chemicals. Nucleic Acids Res 2018; 46: W257–263.

13.

Irwin

Sterling

Mysinger

, et al. ZINC: A free tool to discover chemistry for biology. J Chem Inf Model 2012; 52: 1757–1768.

14.

Kim

Thiessen

Bolton

, et al. PubChem substance and compound databases. Nucleic Acids Res 2016; 44: D1202–D1213.

15.

Weininger

. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. J Chem Inf Comput Sci 1988; 28: 31–36.

16.

Kroes

Renwick

Cheeseman

, et al. Structure-based thresholds of toxicological concern (TTC): Guidance for application to substances present at low levels in the diet. Food Chem Toxicol 2004; 42: 65–83.

17.

Benigni

Bossa

Tcheremenskaia

. In vitro cell transformation assays for an integrated, alternative assessment of carcinogenicity: A data-based analysis. Mutagenesis 2013; 28: 107–116.

18.

Benigni

Bossa

Tcheremenskaia

. Nongenotoxic carcinogenicity of chemicals: Mechanisms of action and early recognition through a new set of structural alerts. Chem Rev 2013; 113: 2940–2957.

19.

Benigni

Bossa

. Mechanisms of chemical carcinogenicity and mutagenicity: A review with implications for predictive toxicology. Chem Rev 2011; 111: 2507–2536.

20.

Benigni

Bossa

Tcheremenskaia

, et al. Development of structural alerts for the in vivo micronucleus assay in rodents. JRC Scientific and Technical Reports. EUR 23844 . Luxembourg: Office for Official Publications of the European Communities, 2009, 42 pp.

21.

Enoch

Cronin

. A review of the electrophilic reaction chemistry involved in covalent DNA binding. Crit Rev Toxicol 2010; 40: 728–748.

22.

Enoch

Ellison

Schultz

, et al. A review of the electrophilic reaction chemistry involved in covalent protein binding relevant to toxicity. Crit Rev Toxicol 2011; 41: 783–802.

23.

Drwal

Banerjee

Dunkel

, et al. ProTox: A web server for the in silico prediction of rodent oral toxicity. Nucleic Acids Res 2014; 42: W53–W58.

24.

Freitas de Freitas

Roth

Abbadi

, et al. Identification of potential inhibitors of Mycobacterium tuberculosis shikimate kinase: Molecular docking, in silico toxicity and in vitro experiments. J Comput Aided Mol Des 2023; 37: 117–128.

25.

Dallakyan

Olson

. Small-molecule library screening by docking with PyRx. Methods Mol Biol 2015; 1263: 243–250.

26.

Fuhrmann

Rurainski

Lenhof

H-P

, et al. A new Lamarckian genetic algorithm for flexible ligand-receptor docking. J Comput Chem 2010; 31: 1911–1918.

27.

Schwaller

Petraglia

Zullo

, et al. Predicting retrosynthetic pathways using a combined linguistic model and hyper-graph exploration strategy. Chem Sci 2020; 11: 3316–3325.

28.

Carnaroglio

Martina

Palmisano

, et al. One-pot sequential synthesis of isocyanates and urea derivatives via a microwave-assisted Staudinger–aza-Wittig reaction. Beilstein J Org Chem 2013; 9: 2378–2386.

29.

Pandey

Dubey

, et al. Mutual amide prodrug of etodolac-glucosamine: Synthesis, characterization and pharmacological screening. Indian J Pharm Sci 2013; 75: 406.

30.

Elangovan

Neumann

Sortais

, et al. Efficient and selective N-alkylation of amines with alcohols catalysed by manganese pincer complexes. Nat Commun 2016; 7:12641.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.11 MB

Toxicological Profiling of Potential Shikimate Kinase Inhibitors Against Mycobacterium tuberculosis

Abstract

Keywords

Introduction

Methods

Identification of possible lead molecules

Prediction of pharmacokinetic properties and drug-likeness by using SwissADME

The prediction of toxicological potential and threat aversion by using ToxTree

Kroes TTC (Threshold for Toxicological Concern)

Carcinogenicity assessment (genotoxicity, non-genotoxicity, mutagenicity)

In vitro mutagenicity assay (Ames Assay) alerts, as per the Istituto Superiore di Sanità (ISS) rule base

Structural alerts for the micronucleus assay in rodents

DNA binding alerts

Protein binding alerts

Prediction of endocrine disruption potential

Prediction of LD50 values

Molecular docking analyses and molecular dynamic simulations

Selection and processing of the macromolecule

Retrieval of ligands

Molecular docking analyses

Molecular dynamic simulations

Results and discussion

Prediction of pharmacokinetic properties and drug-likeness by using SwissADME

Prediction of toxicological potential and threat aversion by using ToxTREE

Prediction of endocrine disruption potential and hepatotoxicity

Prediction of LD50

Molecular docking analyses

Molecular dynamic simulations

Synthesis scheme for two of the proposed ligands

Conclusions

Supplemental Material

Supplemental Material - Toxicological Profiling of Potential Shikimate Kinase Inhibitors Against Mycobacterium tuberculosis

Supplemental Material

Supplemental Material

Supplemental Material

Footnotes

Acknowledgment

Author contributions

Declaration of conflicting interests

Funding

Supplemental Material

References

Supplementary Material

Prediction of LD₅₀ values

Prediction of LD₅₀