A total of 885 African natural products were retrieved from AfroDB (Ntie-Kang et al., 2013) in “.sdf” format for absorption, distribution, metabolism, excretion, and toxicity (ADMET) analysis. The compounds were screened via FAF-Drugs4 (Lagorce et al., 2017) to elucidate their pharmacokinetic, structural, and physicochemical properties. The physicochemical filter used for the compounds was “Drug likeness”. The compounds were categorized as “rejected”, “intermediate”, or “accepted” by FAF-Drugs4. “Accepted” compounds were then chosen for molecular docking.

The “accepted” compounds from the FAF-Drugs4 ADMETox prediction were docked against PfTMPK, using AutoDock Vina (Trott and Olson, 2010) integrated with PyRx (Dallakyan and Olson, 2015). The substrate of PfTMPK (TMP) was also extracted from the complex (Protein Data Bank (PDB) ID: 2wwf) and added to the library of “accepted” compounds for docking. Before the docking, the compounds were energy minimized with OpenBabel using the universal force field (uff) before converting to “.pdbqt” formats (Kwofie et al., 2021). A grid box size of 17.3, 9.7, and 11.2 Å and center dimensions of 42.5, 46.7, and 47.6 Å all in the x, y, and z coordinate axes were used, respectively.

For validation of the performance of the docking technique, the SMILES of five potent inhibitors (compounds 28, 30, 53, 54, and 55) against PfTMPK (Cui et al., 2012) were used to generate their decoys via RApid DEcoy Retriever (RADER) (Wang et al., 2017). A total of 243 decoys were obtained and screened together with the five inhibitors against PfTMPK using AutoDock Vina. The docking results were used to generate a receiver operating characteristic (ROC) curve, and the area under the curve (AUC) was computed using easy ROC Ver. 1.3 (Goksuluk et al., 2016). Parameters used for the ROC generation and calculation of AUC were a non-parametric method for curve fitting (DeLong et al., 1988) for SE estimation and CI as well as a Type I error of 0.05.

The co-crystallized ligands of PfTMPK were removed from their binding site and re-docked against PfTMPK. The best-predicted docking pose of each ligand was superimposed with its respective experimental co-crystallized ligand pose, and their root mean square deviation (RMSD) was calculated using LigAlign (Heifets and Lilien, 2010).

Molecular dynamics simulations were performed using the PfTMPK in complex with TMP and the respective chosen potential lead compounds. GROningen MAchine for Chemical Simulation (GROMACS) version 5.1.4 (Van Der Spoel et al., 2005) was used to perform the molecular dynamics simulations using the GROMOS96 43A1 force field. The topology of the compounds was generated using PRODRG (Schüttelkopf and Van Aalten, 2004). As part of the preparation before the simulation, the complexes were first solvated in a 1-nm dodecahedron water box and neutralized by adding one positive ion. The complexes were relaxed through energy minimization to remove any steric clashes or bad geometry and equilibrated to the required temperature (300K) and density (1,020 kg/m³). After the respective systems were equilibrated and set in the desired temperature and density, a 100-ns production run was performed, and the results of the simulations were analyzed using Xmgrace version 5.1.25 (Vaught, 1996).

One of the potential lead compounds, aurantiamide acetate, was tested for anti-plasmodial activity using the SYBR Green I fluorescence assay as described previously (Smilkstein et al., 2004). Stock concentrations of 100 mM [100% dimethyl sulfoxide (DMSO)] of compounds were diluted with culture media to a working concentration of 100 µM (0.1% DMSO). A serial dilution of 1:2 concentrations of the compound (10 to 0.781 µM) was prepared for the assay. Artesunate was the reference drug diluted from 100 to 3.125 nM. Test wells were initially seeded with 90 µl of ring-stage (synchronized) parasitized red blood cells (pRBCs), culture media, and pRBCs at 2% hematocrit and 1% parasitemia. An aliquot of 10 µl of each concentration was dispensed into each well in triplicates. The wells containing RBCs (2% hematocrit), pRBCs, and cultured protoplast washing (CPW) media served as negative and blank controls and were then incubated for 48 h. An aliquot of 100 µl of 4× buffered SYBR Green I (0.20 µl of 10,000× SYBR Green I/ml of 1× phosphate-buffered saline) was then added for a further 30 min at 37°C. The presence and amount of pRBCs were detected by fluorescence using the Guava EasyCyte HT FACS machine (Millipore, Billerica, MA, USA), and parasitemia was recorded in percentages. The IC₅₀ was extrapolated from non-linear regression curves of percentage inhibition versus log-concentration curves from GraphPad Prism (Graph Pad Software, San Diego, CA, USA) using algorithms obtained from flow cytometry (fluorescence-activated cell sorting (FACS)) data. The experiment was repeated using an asynchronized parasite culture using the same protocol.

The three-dimensional structure of PfTMPK retrieved from the PDB (PDB ID: 2wwf) was solved using X-ray crystallography at a resolution of 1.89 Å (Whittingham et al., 2010). The 2wwf is a homo-oligomer (homodimer) consisting of three chains of PfTMPK in complex adenosine diphosphate (ADP) and thymidine monophosphate (TMP), which is a natural substrate. Other ligands like sodium-ion and glycerol are also present. Figure 1 shows chain A of 2wwf with the active site, and TMP is highlighted. The protein has a sequence length of 212 and is significantly different from the human TMPK with a sequence identity of 36.9%.

ADMET tests are performed to eliminate compounds that may be weak drug candidates so that potential drug-like compounds are prioritized (Sliwoski et al., 2013). FAF drug server uses quantitative QSAR models to predict specific properties or toxicological endpoints of compounds (Lagorce et al., 2017). The predicted properties were compared with the standard range of accepted values concerning the chosen physicochemical filter, which was “drug-likeness”. The filtered compounds were categorized as “rejected”, “intermediate”, or “accepted”. The “accepted” compounds are those with no structural alerts and satisfy the physicochemical filter. Intermediate and rejected compounds show some structural alerts and do not satisfy completely the physicochemical filter. Structural alerts are substructures that are related to mutagenic and carcinogenic properties, which are undesirable for drug-likeness (Benigni and Bossa, 2006). After the compounds were screened using the FAF drug, 91 compounds categorized as “accepted” were used for docking against PfTMPK.

To validate AutoDock Vina’s ability to distinguish between active and inactive compounds concerning PfTMPK, a ROC curve ( Figure 2 ) was generated after virtually screening five potent inhibitors of PfTMPK (Cui et al., 2012) ( Table 1 ) and their decoys against the receptor. The AUC value was calculated to assess the docking performance. AUC value less than 0.5 is considered a poor discrimination ability, from 0.5 to 0.7 is considered moderate, and greater than 0.7 is acceptable (Mandrekar, 2010). AUC value very close to 1 indicates an excellent discriminatory ability of the docking model for the receptor (Mandrekar, 2010). The AUC obtained was 0.95 with a p-value of 9.776159e−78, indicating an excellent discriminatory ability of AutoDock Vina to distinguish between active compounds and decoys of PfTMPK.

To validate further the docking approach used, four co-crystallized PfTMPK ligands (three of which are complexed with inhibitors and a substrate) were retrieved from their PDB crystal structures and re-docked against the PfTMPK protein structures. The co-crystallized binding poses of the ligands were superimposed to their respective predicted docking poses, and RMSDs less than 2 Å were obtained for all of them. This suggests that AutoDock Vina reasonably distinguished between active and inactive compounds for PfTMPK and accurately predicted binding poses.

Ninety-one pre-filtered drug-like compounds were screened against PfTMPK together with its natural substrate and known potent inhibitors. After docking, thirteen of the drug-like compounds had binding affinities greater than the protein’s substrate and were selected as hits ( Table 2 ). The binding energies of three potent inhibitors were −8.9, −9.3, and −9.7 kcal/mol for compounds 28, 53, and 55, respectively, which fell within the same range as the hits. The lowest energies among them were −10.5 and −10.2 kcal/mol for compounds 54 and 30, respectively, which were not far from the lowest among the hits (−9.9 kcal/mol). This is a good indication of the potential antimalarial activity of the hits. Lead compounds were then selected among the hits. Four structural properties of a molecule that ensure proper binding at the active site and inhibitory effect against PfTMPK, developed using QSAR analysis, pharmacophore modeling, and docking studies with a set of thymidine analogs that have well-defined PfTMPK inhibitory activity, were considered in selecting possible lead compounds (Ojha and Roy, 2013). These properties are the presence of –NH fragment, –OH group, urea moiety, and a considerable amount of oxygen atoms (Ojha and Roy, 2013).

The molecular structures of the hit compounds were analyzed to identify any of the aforementioned structural properties. ZINC13374323 possesses –NH group, urea moiety, and a considerable amount of oxygen atoms, while ZINC13365918 also has all the properties but lacks oxygen atoms. ZINC14644461, ZINC95486293, ZINC95486295, and ZINC87493012 all have –OH groups and considerable amounts of oxygen atoms, while ZINC13282986 possesses none of the structural properties. The rest of the six compounds only possess oxygen atoms among the aforementioned structural properties. Most of the compounds seem to possess considerable amounts of O-atoms, which is an essential requirement for high binding affinity toward PfTMPK. The most distinctive structural properties were the presence of NH fragment and urea moiety, which were exhibited only by ZINC13374323 and ZINC13365918. Although all four properties are essential for good inhibitory activity against PfTMPK (Ojha and Roy, 2013), a ligand with at least three of the properties, placing priority on the distinctive properties (presence of -NH fragment and urea moiety), may exhibit a good inhibitory effect against the compound with no or little modifications. As such, ZINC13374323 and ZINC13365918 were considered the potential lead compounds among the 13 hits.

Protein–ligand molecular interactions of ZINC13374323 and ZINC13365918 were further studied and compared to the protein–ligand molecular interaction of compound 25, a PfTMPK inhibitor thymidine analog, which has been shown to exhibit very high activity against the protein (Ojha and Roy, 2013). Compound 25 has been reported to share molecular interactions with protein residues Arg78, Arg99, Arg47, Asp17, Ser22, Phe74, and Tyr43 (Ojha and Roy, 2013). In Figure 3A , we show the substrate to share interactions with similar residues such as Arg78, Arg99, Arg47, and Phe74. From Figure 3C , ZINC13374323 is also shown to have interactions with common residues such as compound 25. ZINC13365918, on the other hand, exhibited interactions with Asp17 and Phe74 among the listed residues ( Figure 3B ). This warranted the experimental characterization of ZINC13374323 to corroborate potential inhibitory activity against PfTMPK. H-bond interactions with Asp17 and Pi-Pi stacked interactions with Tyr107 and Phe74 are common among both ligands. This may indicate the importance of the residues for effective binding. One can suggest that the lack of more O-atoms in ZINC13365918 significantly affects its ability to form H-bonds with important residues such as Arg47 and Arg99. TMP and ZINC13374323 were able to form H-bonds with both Arg47 and Arg99 using their O-atoms. However, a lack of an OH-group did not seem to affect the ligands’ interactions, since none of them formed any bonds with it.

The common name of ZINC13374323 is aurantiamide acetate, which has been shown to exhibit anti-inflammatory and antiviral activities in influenza-infected cells (Zhou et al., 2017). The compound is a constituent of herbs such as the bark of Albizia adianthifolia and Brillantaisia lamium, where it has been shown to exhibit antimicrobial activity (Tamokou et al., 2011; Tamokou et al., 2012). Interestingly, aurantiamide acetate is also an ingredient of A. annua, the Chinese herb from which artemisinin was discovered (Milne et al., 2018). ZINC13365918 is a pyrimidine analog with the name N-hydroxyannomontine. N-Hydroxyannomontine has been shown to exhibit antileishmanial activity against Leishmania braziliensis and Leishmania guyanensis (Costa et al., 2006). The toxicity profiles of both compounds as shown in Table 3 indicate the compounds to be generally safe, with aurantiamide acetate tending to cause some irritation. The toxicity profile was generated with OSIRIS DataWarrior version 5.5.0 (Sander et al., 2015).

We explored the anti-plasmodial activity of aurantiamide acetate further since it is commercially available. N-Hydroxyannomontine is currently not commercially available; as such, in vitro experimentation for this analog was not carried out. An in vitro parasite growth inhibition assay was prepared for the P. falciparum 3D7 strain using the SYBR Green I fluorescence assay (Smilkstein et al., 2004). Artesunate was used as the positive control. The compounds were added to synchronized cultures of the ring stages, and inhibition of the parasite’s growth was determined through IC₅₀, with the experiment conducted in triplicates. The IC₅₀ values obtained for artesunate were comparable to those reported previously with the average being 23.22 nM (Touré et al., 2008; Quashie et al., 2013). The average IC₅₀ value obtained for aurantiamide acetate was 69.33 μM ( Table 4 ). The inhibition curves of both compounds are shown in Figure 7 . A single screen of aurantiamide acetate was also performed on asynchronized cultures of P. falciparum 3D7 strain. The IC₅₀ value achieved for the asynchronized culture was >100 μM ( Supplementary Figure 1 ), which indicates an increase in the potency of the compound when screened against the ring stages of the parasite. This could imply a clue into the target specificity of the compound but requires further investigation; therefore, a stage-of-action study to assess the potency of the compound against different asexual stages is recommended. The activity of aurantiamide acetate, despite being appreciable for a natural compound, must be further improved through optimization to cater to its lack of O-atoms, which is necessary for PfTMPK inhibition.