Anti-trypanosomatid structure-based drug design – lessons learned from targeting the folate pathway

ABSTRACT Introduction Trypanosomatidic parasitic infections in humans and animals caused by Trypanosoma brucei, Trypanosoma cruzi, and Leishmania species pose a significant health and economic burden in developing countries. There are few effective and accessible treatments for these diseases, and the existing therapies suffer from problems, such as parasite resistance and side effects. Structure-based drug design (SBDD) is one of the strategies that has been applied to discover new compounds targeting trypanosomatid-borne diseases. Areas covered We review the current literature (mostly over the last 5 years, searched in the PubMed database on 11 November 2021) on the application of structure-based drug design approaches to identify new anti-trypanosomatidic compounds that interfere with a validated target biochemical pathway, the trypanosomatid folate pathway. Expert opinion The application of structure-based drug design approaches to perturb the trypanosomatid folate pathway has successfully provided many new inhibitors with good selectivity profiles, most of which are natural products or their derivatives or have scaffolds of known drugs. However, the inhibitory effect against the target protein(s) often does not translate to anti-parasitic activity. Further progress is hampered by our incomplete understanding of parasite biology and biochemistry, which is necessary to complement SBDD in a multiparameter optimization approach to discovering selective anti-parasitic drugs.


Introduction
Trypanosomatids cause devastating life-threatening diseases in humans and animals that are transmitted through insects. These include sleeping sickness -caused by Trypanosoma brucei, Chagas disease -caused by Trypanosoma cruzi and leishmaniases -caused by Leishmania species. Cases are mostly observed in developing countries in tropical regions of the world, but are also seen in Southern Europe [1][2][3]. Currently, 6-7 million people, mostly in Latin America, are estimated to be infected with Chagas disease [2]. Leishmaniases are endemic in 92 countries, and currently it is estimated that 700,000 to 1 million cases occur annually [3]. In contrast, the number of recorded cases of sleeping sickness (Human African Trypanosomiasis, HAT), is systematically decreasing, and fell below 1000 in 2019 [1], which was supported by the availability of the new eflornithine-nifurtimox combination therapy and the recently approved oral drug fexinidazole [4][5][6][7], illustrating the impact and importance of developing drugs against trypanosomatids. However, even with such success, the recurrence of epidemics can never be excluded as, for example, seen for malaria in India in the 1960s [8]. In particular, drug resistance and the existence of parasite reservoirs in animals may pose future challenges for existing therapies [9][10][11][12]. For African trypanosomiasis, one of the greatest problems is the animal reservoirs, such as in cattle. Even if HAT is combated, the more limited management of trypanosomiasis in domestic animals, which have been also shown to be a reservoir of human pathogens, may still hamper the overall success of HAT management [13][14][15].
Trypanosomatids are effective in evading host immunity and have complicated life cycles involving many distinct stages (Figure 1), and many details of their biochemistry differ from that of other eukaryotes and are not well known [16,17]. So far, no vaccines are available against trypanosomatid-borne diseases. Moreover, few treatment options are available, and they are often not effective enough, have significant side effects, or are ineffective due to parasite resistance. For example, miltefosine, which is used to treat leishmaniasis, is teratogenic [18] and resistant strains to drugs such as pentamidine, used against T. brucei [19], or anti-leishmanial antimonials [20] are known.
Therefore, even in the best-controlled case of sleeping sickness, better treatments, that overcome resistance and have reduced side effects, are needed. Unfortunately, drug design efforts against trypanosomiases are not generally profitable for pharmaceutical companies, since these so-called 'neglected diseases' occur mostly in poor regions of the world [21]. However, in recent years, there have been several initiatives to advance drug design against neglected tropical diseases. These include (i) the Drugs for Neglected Diseases initiative (DNDi, https://dndi.org [22]), (ii) the Trypanogen and Trypanogen+ projects funded by AAS/Wellcome under the H3Africa initiative (http://trypanogen.net/ [23]), (iii) two EUfunded projects that focused on targeting specific biochemical pathways of parasites causing the diseases: New Medicines for Trypanosomatidic Infections (NMTrypI [24], https://fp7nmtrypi.eu/ [25], https://cordis.europa.eu/project/id/603240 [26]) and Parasite-specific cyclic nucleotide phosphodiesterase inhibitors to target Neglected Parasitic Diseases (PDE4NPD, https://cordis.europa.eu/project/id/602666 [27]). One of the focuses of the NMTrypI project, in which the authors of the present article participated, was targeting the parasite folate pathways. We here review recent efforts in antitrypanosomatid structure-based drug design (SBDD) from this perspective.
SBDD strategies are employed in many anti-trypanosomatid drug design campaigns to guide and interpret experiments [28]. In recent years, efforts have been made to identify valid antitrypanosomatid targets [29]. The enzymes and biochemical pathways most frequently targeted by SBDD include (i) the folate pathway, (ii) lanosterol 14α-demethylase (CYP51), (iii) phosphodiesterases, (iv) cysteine proteases, and (v) the trypanothione Article highlights • The trypanosomatid folate pathway is a potential target for antiparasitic drugs. • Many crystal structures have been solved for the key trypanosomatid folate pathway enzymes. • Structure-based drug design has led to multitarget, selective enzyme inhibitors. • Multiple factors must be considered to optimize anti-parasite activity. • Improved modeling techniques and better knowledge of the biology of parasites should enable the discovery of more potent selective antiparasitic agents.
This box summarizes key points contained in the article. metabolism pathway (for review, see ref [30]). Here, we focus on targeting the folate metabolism of trypanosomatid parasites. For reviews on anti-trypanosomatid SBDD against other targets, see also ref [31]. Computational approaches against trypanosomiases have also been reviewed by Pereira et al. [32].
In the folate pathway (Figure 2), folates are reduced to produce metabolites for DNA synthesis. Blocking the folate pathway is a strategy that has been pursued in anti-cancer, anti-bacterial and anti-malarial drug design [33][34][35]. A similar approach has been taken for anti-trypanosomatid drug design [36][37][38]. Whereas in the former cases, this strategy requires inhibiting dihydrofolate reductase (DHFR), for trypanosomatids, apart from DHFR (in the bifunctional DHFR-TS enzyme), pteridine reductase 1 (PTR1) also needs to be inhibited ( Figure 2). PTR1 can reduce pterins and folate, and when DHFR is inhibited, it can act as a bypass enzyme and retain folate reduction [39,40]. Notably, PTR1 is not present in humans and therefore targeting PTR1 provides a way to specifically target the parasite folate pathway while minimizing side-effects. The recent applications of SBDD against the trypanosomatid folate pathway discussed here focus mostly on PTR1, with parasite DHFR (from DHFR-TS) often being treated as an additional target. Parasite TS was rarely targeted because the sequence similarity of this enzyme in trypanosomatids to the human homologue is very high, hindering the design of selective antiparasitic compounds [41].
T. brucei PTR1 (TbPTR1) was confirmed by gene knockout experiments to be a drug target in its own right [42,43]. T. cruzi PTR1 (TcPTR1) overexpression was also shown to result in methotrexate (MTX, Table 1) resistance in T. cruzi, but the enzyme was detected only in epimastigote (insect) stage and not in amastigote (intracellular) or trypomastigote (bloodstream) stages [38]. LIkely due to the gaps in knowledge of PTR1 expression levels, there are few SBDD studies focusing on the T. cruzi folate pathway, despite T. cruzi causing one of the least managed trypanosomatidic infections, especially when considering chronic disease [54]. The TcPTR1 structure has not been solved, but Senkovich et al. have solved a structure of a close paralog, TcPTR2 [55]. Another reason for the limited targeting of the T. cruzi and Leishmania folate pathway enzymes for drug discovery may be related to their druggability: From a comparative analysis of sequence and structure-based features [41], it was observed that TbPTR1 is more druggable than leishmanial PTR1 and TcPTR1. Therefore, drug design attempts focusing on inhibiting TbPTR1 have mostly been more successful than for Leishmania and T. cruzi PTR1.
The natural starting points for SBDD to block the trypanosomatid folate pathway are known inhibitors, such as the anticancer drugs MTX and pemetrexed, or anti-microbial drugs such as trimetrexate [48,56,57] (Table 1). Such compounds generally display lower activities against PTR1 than DHFR. MTX, for example, inhibits human DHFR (hDHFR) at the subnanomolar level, while it is a submicromolar inhibitor of PTR1 [39,44]. Pteridine derivatives have been widely studied as The key folate pathway enzymes in trypanosomatids and the catalyzed reactions (arrow color corresponds to the color of the enzyme structure; the dotted line indicates that the DHFR activity toward folate is low when compared to dihydrofolate [37]). The enzyme structures are shown for T. cruzi DHFR-TS (PDB code: 3INV), and a monomer from the modeled homotetramer of T. cruzi [41]; c) Main on-targets and off-targets that need to be considered for inhibiting the trypanosomatid folate pathway, schematically shown in their cellular location (in Leishmania promastigotes that invade macrophages). The figure is adapted from ref [41]. with permission from the copyright holder.
A variety of SBDD approaches and combinations thereof have been used so far (Tab. S1, Supplementary Information). These mostly include crystallography in combination with homology modeling, molecular docking, VS of compound libraries, and molecular dynamics (MD) simulations. Here, we present recent applications of SBDD and describe how they have yielded compounds with on-target and on-parasite activities while considering challenges for drug design, such as (i) off-target selectivity, (ii) targeting multiple enzymes, (iii) translation from predicted and experimental on-target to on-parasite activities.
This review covers the current literature on SBDD against the trypanosomatid folate pathway. Most of the research articles could be found in the PubMed database [60] on 11 November 2021 with the following search term: '(trypanosoma or leishmania) and (folate or pteridine or dhfr or ptr1 or thymidylate) and (qsar or pharmacophore or 'molecular dynamics" or docking or machine or virtual or computational or modeling or modeling)" and limiting the results to the 5 years before the aforementioned date.

Considering multiple targets
Targeting the trypanosomatid folate pathway generally requires simultaneous inhibition of DHFR as well as the PTR1 bypass, although PTR1 may be suitable as the sole target in T. brucei (see, e.g. [61][62][63][64]). Furthermore, the homologous hDHFR enzyme (with about 30% sequence identity to parasite DHFR and a similar pocket [41]) needs to be considered as an off-target when designing anti-trypanosomatid DHFRtargeting inhibitors. This makes the antifolate design process quite complex. Panecka-Hofman et al. therefore devised a computational workflow for comparative analysis of the trypanosomatid and human folate pathway targets for the design of selective antiparasitic agents [41,65]. This workflow highlighted some overlapping properties of the respective parasitic DHFR and PTR1 pockets that could be exploited in multitarget ligands but also showed that it is difficult to avoid such ligands binding to the hDHFR off-target. We think that such on-and off-target selectivity should be considered early in the drug discovery process during the selection of the biochemical pathway and enzymes targeted, and in deciding whether to aim for a single compound to bind to multiple targets or several compounds with different targets that could be administered simultaneously.
Several recent studies have considered PTR1 and DHFR in multi-target ligand design approaches. Cavazzuti et al. [44] studied pteridine derivatives considering the PTR1, TS, and DHFR enzymes of L. major and T. cruzi. Many compounds with other non-classical scaffolds, such as thiadiazoles, benzothiazoles and flavonoids investigated in the NMTrypI project, have been tested against PTR1 and DHFR [66][67][68], but were mostly ineffective against the latter. On the other hand, Poehner et al. targeted both TbPTR1 and TbDHFR, achieving the most potent known inhibitors of TbPTR1 (pteridine Table 1. Compound structures, selected on-and off-target (PTR1, DHFR) activities and other information for the reference antifolates. derivatives with apparent inhibition constants in the picomolar range) that also selectively targeted DHFR [59]. Another potential interesting starting point to target both TbPTR1 and TbDHFR is pyrimethamine (Table 1), which has submicromolar activities against both enzymes [45]. Herrera-Acevedo et al. evaluated a kaurane library against multiple leishmanial PTR1 enzymes. The selected compounds predicted by molecular docking to be good binders of L. major PTR1 (LmPTR1) and by a machine-learning model to be active against L. major, were also evaluated in silico to favorably bind L. panamensis, L. amazonensis and L. braziliensis PTR1 (LbPTR1) [50], which has, however, not yet been verified experimentally. However, in general, PTR1 targets from different pathogenic trypanosomatid species have significantly dissimilar pockets [41] (Fig. S1, Supplementary Information). Therefore, the ambitious task of designing a broad-spectrum inhibitor targeting the main trypanosomatid pathogens -Leishmania species, T. brucei, T. cruzi -via the folate pathway, might be difficult. This problem is underscored by the fact that few inhibitors targeting PTR1 enzymes from more than one species were identified so far. When the PTR1 (and sometimes DHFR) enzymes of Leishmania and T. brucei species were targeted together, leishmanial targets were generally more weakly inhibited than those from T. brucei (e.g [59,68,69]), which reflects the lower druggability of the former.
The first structures of complexes with PTR1 and DHFR were determined for classical pteridine inhibitors [39,57,76,77]. The structures revealed that antifolates adopt similar binding modes to the folate substrate in PTR1 (Figure 3(a)). In this binding mode, the compounds make extensive interactions with the NADP+/NADPH cofactor (sandwich stacking involving nicotinamide, hydrogen bonds or salt bridges with phosphates and ribose, see Figure 3), which must be considered in computational studies. At the other end of the binding site pocket, the compounds often interact with the C-terminal tail of another subunit of the PTR1 homotetramer (Figure 3(a)). This tail contains a basic residue -His in TbPTR1 (pos. 267) or Arg in LmPTR1/TcPTR1 (see the alignment in Fig. S1, Supplementary Information). Therefore, at least the two terminal residues from the other chain (e.g. in TbPTR1: His267-Ala268, see Figure 3(a)) or the full PTR1 homotetramer should be considered in modeling studies.
More recently, crystallography has been used to reveal the binding modes of non-pteridine scaffolds, including 2-amino-1,3,4-thiadiazoles [67], 2-aminobenzothiazoles [68], and flavonoids [66,72], which display the 'classical' binding mode with an aromatic ring system bound in the PTR1 biopterin/pteridine pocket (Figure 3(b-d) and Table 2). Recent crystal structures of TbPTR1 with the first tricyclic-based compound (compound 1) [71] (Table 2) and with pyrimethamine ( Table 1) [45] indicated further potential new starting points for drug design. The only compounds found to have nonsubstrate-like binding modes in alternative subpockets of the TbPTR1 binding site were developed by Mpamhanga et al. [61]; the structure of one of these, compound A1, bound to TbPTR1 is shown in Figure 3(e).
Structural alignment of the PTR1 crystal structures [41] shows that the so-called substrate loop, which flanks the active site, adopts a range of conformations in the crystal structures. These affect the size of the binding site and, in TbPTR1, the position of Trp221, which lines one of the subpockets in the vicinity of the active site. However, the substrate loop (and the active site) dynamics might be greater in solution in the real biological environment than in the crystals, due to crystal packing interactions that could limit the movements of the loops. The binding site dynamics pose challenges for the prediction of the binding modes of some compounds, even using induced-fit docking protocols, and in some cases, it may be necessary to apply ensemble docking or MD simulations. The dynamic character of the substrate loop and the effect on ligand binding have rarely been accounted for in SBDD, despite several studies involving MD simulations (discussed later).
An interesting observation is the partial ligand occupancy of the four binding sites in the PTR1 homotetramer crystal structures. In some structures, only three of the four sites are occupied by ligands; e.g. by flavonoids bound to TbPTR1 (PDB: 5JCJ, 5JCX, 5JDI; e.g. compound 2 displayed in Figure 3(c,d) and Table 2) [66]. This indicates that subtle effects of binding cooperativity in the PTR1 homotetramer and long-range dynamics may affect ligand binding and PTR1 inhibition.
Since the PTR1 active site (like its substrates) is relatively polar [41], water molecules (see, e.g. Figure 3(a-d)) and protonation states play a significant role in binding, with implications for docking workflows. For example, an NMR study by Cocco et al. showed that at pHs up to 9.2, MTX binds hDHFR with the N1 nitrogen of the pteridine in the protonated state ( Figure 3a) interacting with the negatively charged Glu30 in the active site [86], whereas the pKa of N1 is about 5.73 in the unbound state of MTX [87]. A similar pKa shift will likely occur when MTX binds parasitic DHFR and PTR1, due to the similar, negatively charged PTR1 pocket lined by phosphates of the NADP cofactor (Figure 3(a)). Notably, the N3 nitrogen of folate, located in the same place as the N1 of MTX in superimposed structures, is protonated in the neutral state and makes a hydrogen bond with the NADP phosphate oxygen (Figure 3(a)). Depending on the pH, Asp161 of TbPTR1 ( Figure 3, 181 in LmPTR1 and 169 in TcPTR1), likely involved in the enzymatic catalysis mechanism [76], and His267 of TbPTR1 ( Figure 3) may also adopt different protonation states, which may affect the binding of charged ligands.

VS of natural products
Many VS studies start with natural products [88]. For instance, Morales-Jadán et al. [78] virtually evaluated three alkaloids (aspidoalbine [compound 3] in Table 2, aspidocarpine [A3] and tubotaiwine [A4] in Tab. S3 in Supplementary Information) from the traditional anti-malarial and anti-leishmanial Amazonian medicinal plant Aspidosperma spruceanum against key leishmanial targets: pyruvate kinase, hypoxanthine-guanine phosphoribosyltransferase, squalene synthase, and both DHFR-TS and PTR1. This computational study predicted better affinity for aspidoalbine than some   native PTR1 ligands, though no experimental validation has been provided so far.
In another study, an in-house library of 118 sesquiterpene lactones, plant-derived terpenoids, was investigated [79] in a pharmacophore-based VS approach to find binders of TbPTR1 and TbDHFR. 29 hits were tested in vitro on the enzymes resulting in the identification of cynaropicrin (compound 4, Table 2) with IC 50 against TbPTR1 of 12.4 μM and against TbDHFR of 7.1 μM. Another library of sesquiterpene lactones (dataset in SistematX database: https://sistematx. ufpb.br [89]) was virtually screened against leishmanial targets [74]. Herrera-Acevedo et al. combined structure-based and ligand-based VS approaches. For the latter, compound sets with activities against L. donovani amastigotes and promastigotes in vitro (3159 and 1569 compounds, respectively) were obtained from ChEMBL [90,91] and used as a basis for random forest (RF) models that achieved an accuracy over 71% in filtering for actives. In parallel, docking-based VS of the SistematX library was done against L. donovani proteins: N-myristoyltransferase, ornithine decarboxylase, mitogenactivated protein kinase 3, and a homology model of PTR1. While there was no experimental validation, the combination of the results from both VS approaches pointed to 13 compounds that were predicted to bind to the chosen targets and have antileishmanial properties.
The same combination of ligand-based and structure-based VS was also applied to an in-house library of 360 tetracyclic diterpenes, kauranes [50] (see workflow in Figure 4(a)), to identify 5 LmPTR1-targeting anti-leishmanial molecules that were classified as active by the combined docking and ligandbased (RF model) approaches. IC 50 s against LmPTR1 of three out of the four predicted actives tested were below 10 μM (Table 2, compounds 5, 6, and 7). Finally, molecular docking calculations were extended to additional PTR1 models from L. braziliensis, L. panamensis, and L. amazonensis, revealing the same compounds among the 10 top-ranked compounds. The Abbreviations: 'p' -promastigote, 'a' -amastigote, db -database, NI -no inhibition. a When K i or K i_app values are available, these are given in μM. complexes of three of these with LbPTR1 were additionally tested by MD simulations and MMPBSA calculations. Favorable predicted ADMET properties highlighted these compounds as good starting points for future drug design. In another VS approach, pharmacophore queries created based on four crystal structures were used to screen a library of 1100 commercially available natural products supplied by PhytoLab GmbH & Co. KG (Vestenbergsgreuth, Germany; http://phyproof.phytolab.de/ [92]) against LmPTR1 [81]. Eighteen hits were tested in vitro on recombinant LmPTR1, 15 of which displayed significant inhibitory activities at 50 μM. For the best six compounds, IC 50 values were measured and the most active compound was sophoraflavanone G (compound 8, Table 2) with IC 50 of 19.2 µM, which could be a potential lead for further LmPTR1 inhibitor design.
In another study (Figure 4(b)), Kimuda et al. [82] performed molecular docking of a custom library of 5742 compounds from ZINC [93] and the South African Natural Compounds database (SANCDB) [94,95] against TbPTR1/DHFR to select 18 compounds (all hits present in ZINC), whose binding modes were confirmed by MD simulation. Of these, five compounds with low micromolar activities in vitro against T. brucei were identified (e.g. compounds 9-11 in Table 2). While one compound appeared to be a selective TbPTR1 inhibitor, antagonism of four inhibitors when combined with a known TbDHFR inhibitor suggested that those compounds could act on TbDHFR.
Finally, combining VS against TbPTR1 with parallel experimental in vitro screening of a natural products library [66] established flavonols as promising TbPTR1 binders with antitrypanosomal activity. This led to the synthesis of 16 derivatives, 12 of which had EC 50 values against T. brucei below 10 μM. The chemical structures of two selected compounds (2 and 12) are shown in Table 2.
In summary, recent VS approaches utilizing natural products substantially expanded the chemical space of potential PTR1 inhibitors. The activities of some compounds were tested in vitro in enzymatic assays and reached the low μM range (e.g. flavonols and kauranes [50,66]), and in a few studies onparasite activities were also measured and showed active compounds [66,79]. An interesting strategy to increase the likelihood of identifying compounds with antiparasitic activity in the absence of assays was presented by Herrera-Acevedo et al. in their combination of structure-based screening with ligand-based classification models trained on activity data in the hit selection [50,74].

Known drugs as starting points and drug repositioning
Drug repositioning is a strategy to identify compounds against a new target that already have proven favorable pharmacokinetic and ADMET properties, thus having a better chance to shorten the drug development process [96]. This strategy is also used in drug design against neglected diseases [97]. For instance, since the initial idea of targeting the trypanosomatid folate metabolism was transferred from blocking the folate metabolism in cancer and microbial infections, several studies started from folate-like drugs like MTX or trimetrexate (Table 1), e.g [44,77]. Early VS studies by Ferrari et al. [83] revealed riluzole (compound 13, Table 2) -a CNS drug -as a hit. The activity of riluzole was further confirmed on Leishmania mexicana and L. major, especially during their exponential growth phase. It was further demonstrated that after treatment with riluzole, parasite cells had significantly lower PTR1 activities, and were also more susceptible to oxidative stress [84]. A compound series with the benzothiazol-2-amine core of riluzole was later designed and tested against LmPTR1 and TbPTR1 [68]. In this series, one compound 14 (Table 2) with EC 50 of 7 μM against T. brucei bloodstream forms was found to be safe, orally available, and was selected as a potential candidate for further in vivo studies.
Cycloguanil (Table 1), a known antimalarial DHFR-targeting drug (metabolite of proguanil) that also inhibits TbDHFR [53], was more recently found to have modest activity against TbPTR1 (IC 50 of 31.6 μM) [70]. A small library of derivatives was synthesized and tested in vitro on TbPTR1, and the two best compounds (15 and 16, Table 2), showed IC 50 values of 692 and 186 nM, respectively. Compound 16 was furthermore found to be effective against T. brucei in an in vitro cell growth inhibition test, pointing to 1,6-dihydrotriazines as potential starting scaffolds for anti-trypanosomatid antifolate design. In a follow-up study [45], Tassone et al. studied further cycloguanil derivatives and crystallized another DHFR inhibitor, pyrimethamine (Table 1), in the TbPTR1 active site. Comparative analysis of the PTR1 binding modes of these DHFR inhibitors provided insights for the design of dual PTR1/DHFR inhibitors.
Juarez-Saldivar et al. [85] virtually screened 1857 FDAapproved drugs against TcDHFR-TS. To select the top docked poses, they used the Tanimoto similarity between proteinligand interaction fingerprints (PLIFs) of the screened compounds and known inhibitors instead of the docking score. They found known TcDHFR-TS inhibitors to have the poorest docking scores among the screened compounds, but their PLIF-based scores were among the best. They selected 10 drugs with Tanimoto/PLIF-based scores comparable to known inhibitors, as well as more favorable docking scores. They belonged to six divergent classes in terms of chemical structure and medical applications. Three of these drugs gave promising results: nilotinib, glipizide, and gliquidone (compounds 17, 18, 19, Table 2) inhibited T. cruzi epimastigote cell growth in the 10-micromolar range but also displayed cytotoxicity against human foreskin fibroblasts and their effect on other T. cruzi life-cycle stages was not determined.

Combinatorial libraries and fragment-based drug design (FBDD)
FBDD and the construction of combinatorial libraries can substantially expand the chemical space of the active compounds. Poehner et al. [59,65] took a multistep approach to the optimization of pteridine derivatives as inhibitors of kinetoplastid folate pathway enzymes. Starting from biochemical, crystallography-and docking-based evaluation of reference compounds and initial structure-based design, 20 novel inhibitors of T. brucei and L. major PTR1 and the corresponding DHFR enzymes were designed. While many of the new derivatives exhibited subnanomolar PTR1 inhibition and selective submicromolar inhibition of parasite DHFR, their anti-parasitic effect varied. The design of further derivatives with optimized trypanocidal effect was carried out by starting from VS of a combinatorial library composed of fragments of the previously designed compounds (Figure 4(c)). Docking of the library against parasitic enzymes and hDHFR in conjunction with a compound-descriptor-based optimization approach of the on-parasite activity allowed the selection of additional six pteridines, the best three of which (compounds 20, 21, and 22, Table 2) reached low micromolar inhibition of T. brucei. Sharma et al. used pharmacophoric approaches to virtually screen a combinatorial library composed of selected head groups, linkers, and tails with the scaffolds selected based on their proton affinity as determined by quantum mechanics (QM) methods to effectively target the negatively charged LdDHFR active site [98]. Then, docking-based VS was performed, and promising hits were evaluated against the hDHFR to filter out five compounds selectively targeting LdDHFR. The stability of their interactions with LdDHFR was verified by MD.
In another study incorporating QM calculations, Jedwabny et al. [99] explored the computationally efficient approximate QM-based ab initio binding energy model, including only multipole electrostatic and dispersive energy terms [100], to evaluate the relative interaction energies of a small series of fragments (compound substituents) with subpockets of the TbPTR1 active site. They found a significant correlation between the computed energies and IC 50 values for one subpocket (−0.96), better than for the commonly used empirical Glide XP scoring (−0.82) [101]; the model was then transferred to another TbPTR1 subpocket to predict interaction energies of a small series of fragments, demonstrating the applicability of the method for evaluating fragment binding affinities.

Considering receptor flexibility
Structural alignment and comparative analysis of the crystal structures with the TRAPP web server [102] showed the conformational variability of the PTR1 and DHFR binding pocket [41]. In PTR1, the side chains and the substrate loop adapt their conformations depending on the bound ligand [41]; see e.g. the movement of Trp221 in the TbPTR1 pocket shown in Figure 3(f).
Many recent drug design workflows have incorporated MD simulations of ca. 50-100 ns of PTR1 [50,73,74,82,103,104] or parasite DHFR [73,98,105]. These simulations were mostly of a single monomer of PTR1 or the DHFR subunit of the DHFR-TS parasitic enzyme. In most cases, MD was used to validate the binding modes from molecular docking, which were mostly stable on the time scale of the simulations performed, so the simulations did not discriminate binders from non-binders. Longer trajectories of full PTR1 homotetramers or DHFR-TS dimeric enzymes could be considered in the future, though these would be substantially more computationally expensive.
MD simulations also yielded mechanistic insights. For instance, Istanbullu et al. highlighted the importance of watermediated hydrogen bonds for the stabilization of all thiazolopyrimidine ligands in the LmPTR1 binding site [103]. Kimuda et al. [82] observed a lower radius of gyration, corresponding to an increased compactness, of all the simulated TbPTR1 systems with respect to the initial structure and for ligand-bound structures with respect to the unbound TbPTR1 structure. They also noted a higher flexibility of the substrate loop compared to other regions of TbPTR1, and in some of the simulated complexes (e.g. compounds 9-11, Table 2), the substrate loop was more flexible than in the unbound TbPTR1. Principal Components Analysis and analysis of the average betweenness centrality parameter of the residue network indicated differences in information flow upon binding of different compounds. Insights into the long-range dynamics of the unbound LmPTR1 homotetramer in complex with NADP + were also gained by Panecka-Hofman and Wade [106] from normal mode analysis, which showed concerted motions of the four substrate loops, and Rotamerically Induced Perturbation [107] simulations, which detected flexibility hot-spots and possible allosteric couplings between distant PTR1 residues.

Estimating binding affinities
Due to the known deficiencies of docking software in estimating binding free energies (lack of correlation of docking scores with experimental affinities [108]), MMGBSA and MMPBSA methods are often used to estimate the relative affinities of the docked and simulated ligands [50,74,82,98,103,105]. In one of the few studies that provides both the computed binding energies and experimental values [50], the three selected kauranes had lower MMPBSA free energies (compound 5: −132.7, compound 6: −121.4 compound 7: −138.3 kcal/mol, see Table 2) than the reference inhibitor pyrimethamine (−110.0 kcal/mol, Table 1). Notably, although the binding free energies of the compounds were considerably overestimated, the ranking of the activities of the three kauranes predicted by MMPBSA for LbPTR1 was the same as for LmPTR1 IC 50 s (for activities, see Table 2). However, the experimental IC 50 value for pyrimethamine against LmPTR1 (1.1 μM) was significantly lower than for kauranes, in contrast to the MMPBSA calculation for LbPTR1. This might be explained by the very different chemical structures (and thus, likely, physicochemical properties) of pyrimethamine and kauranes (see Tables 1 and 2). In particular, it was found that the van der Waals component of the binding free energies of kauranes is the main determinant for binding, while for pyrimethamine it is the electrostatic component.
Molecular mechanics-based methods to estimate compound affinities are limited by force-field accuracy. Halogen atoms, which are frequently found in drug-like compounds, are often inadequately represented in classical force fields, especially when they form halogen bonds [109]. Therefore, modeling, for example, the TbPTR1 binding of the 2-aminobenzothiazole derivatives developed by Mpamhanga et al. or Spinks et al., and in general -many other drug-like halogenated compounds, represents a challenge [61,63]. The aforementioned ab-initio QM model [100], which proved to accurately predict relative binding energies of a series of compound substituents (including a chlorine atom) binding to the TbPTR1 subpockets, could address this problem, since it is capable of modeling the halogen atom electronic structure [99].

Conclusions
The applications of SBDD to the trypanosomatid folate pathway to date have focused mostly on targeting the main 'bypass' enzyme, PTR1, though in some studies, DHFR has been considered together with PTR1 or as a sole target. These studies targeted the biopterin and folate binding sites of the enzymes. Established state-of-the-art approaches, combining crystallography, enzyme kinetics, and parasite inhibition assays with virtual screening, docking, and MD simulation, have been used.
The most active compounds, with the greatest potential for multitarget inhibition, are pteridine derivatives. Drug repurposing campaigns have also been quite successful, while screening of natural product libraries is so far at the stage of hit identification and will need further work to identify validated leads for lead optimization.
Much structural data on PTR1 are now available which may serve as a strong basis for further SBDD against the trypanosomatid folate pathway. The flexibility of the proteins is increasingly considered in SBDD by MD simulation. It is worth noting that missing biological data and knowledge of parasite mechanisms are major factors hampering the progress in anti-trypanosomatid drug design (in contrast, for example, to the human folate pathway). Moreover, drug discovery pipelines that integrate SBDD with the multiple factors relevant for drug efficacy against parasites are required.

Summary of compounds -successes and challenges
In the recent attempts to inhibit the PTR1 and DHFR enzymes, the compound classes explored included pteridines [59], as derivatives of the natural substrates, and compounds with bioisosteric moieties, like 2-amino-1,3,4-thiadiazoles [67] and benzothiazoles [68]. The natural products VS results included diverse compound classes, such as flavonoids, kauranes, and sesquiterpene lactones [50,66,74,79], and several known drugs were repurposed as trypanosomatid folate pathway inhibitors [70,85]. Notably, the first tricyclic core targeting the TbPTR1 pocket was developed [71], which could be an interesting starting point for a more potent and selective PTR1-targeting compound series. Most of the designed compounds and VS hits were, however, more active against TbPTR1 than LmPTR1, and on T. brucei parasites than Leishmania parasites (e.g [59,68,69]).
The most potent series (reaching subnanomolar TbPTR1 inhibition) with some potential to reach multiple targets due to their similarity to the natural substrates, were the recently explored pteridines [59]. In this series, selectivities expressed as IC 50 ratios reached 169 for TbDHFR vs. hDHFR. In contrast, the PTR1-targeted non-pteridine compound series (thiadiazoles [67], benzothiazoles [68], flavonoids [66]), though having other favorable properties, often lost their potential to bind parasitic DHFR. Unfortunately, despite nanomolar or even subnanomolar on-target activities, so far no compounds have been developed further as single anti-trypanosomatid agents, though some promising, potentially useful, candidates were selected, e.g. compound 14 of the 2-aminobenzothiazole series [68]. However, multi-compound therapy may be considered, e.g. several compound series were also tested as potentiation agents in combination with the DHFR inhibitor MTX [66][67][68], but ultimately, a selective inhibitor of parasite DHFR is an unmet need.

Translation of on-target activities to on-parasite activities and in vivo efficacies
A problem that has been encountered in many of the SBDD studies is insufficient translation of on-target activities to significant, low micromolar, on-parasite activities. In earlier studies, the problem was analyzed by Spinks et al. [63] and the main hypothesis, apart from possible negative cooperativity of inhibitor binding to PTR1 and probable sequestration of lipophilic compounds in the membrane, was that for compounds targeting TbPTR1 to be effective, they should display subnanomolar potencies due to the low K m value for this enzyme, which was previously overestimated.
One of the reasons why inhibitors that are active against enzyme targets do not exert the required activities on parasite are transport-related issues: drugs must reach the parasite cell or its compartments and, for some developmental stages, cells of the mammalian host, like macrophages, in which promastigotes of Leishmania reside (Figure 1). In some cases, considering specific transporters that allow for crossing cell membranes could be a useful option. This, for example, applies to the polar pteridines, which, among others, may use the host folate receptor β transporter [41]. Another approach to tackle this problem was presented by Poehner et al., who proposed compound property-based optimization together with a docking-based approach to filter compounds for improved anti-parasitic activities [59].
Another important consideration for any antiparasitic compound is its safety for the host organism. To assess compound toxicity, in some studies, compounds were tested against selected cell lines, see, e.g. [82,85,103]. In most studies arising from the NMTrypI project [59,[66][67][68]72], an extensive early ADME and toxicity panel was used, including assays against the hERG potassium channel, aurora B kinase, cytochrome P450s, as well as mitochondrial toxicity and cytotoxicity assays. Such assays help to select safe compounds for in vitro and in vivo pharmacokinetic studies [66,68]. However, follow-up studies would be necessary to pinpoint and optimize compound's profile in the host context before progressing to more in-depth in vivo studies.
Furthermore, novel delivery mechanisms, such as nanoparticles, could be used to circumvent some of the transport and toxicity-related issues. This strategy has shown very promising outcomes, including increased efficacy and reduced toxicity of anti-trypanosomatid drugs in in vitro studies and animal models, as, e.g. reviewed in refs [110][111][112]. One example in the field of targeting the trypanosomatid folate pathway was encapsulating a flavonoid derivative with either hydroxypropyl-β-cyclodextrins or poly(lactic-co-glycolic acid nanoparticles [66]: this, however, did not increase drug concentration in mouse blood plasma (the primary aim), but distinctly reduced THP1 toxicity. In summary, these types of approaches could solve some transport and toxicity issues for the designed anti-trypanosomatid antifolates.
Finally, targeting the parasites is even more difficult due to their complicated life cycle (Figure 1), where one must consider multiple development stages with differing barriers and potential drug evasion mechanisms, which include, e.g. the variant surface glycoprotein (VSG) coat that contributes to antigenic variation of trypanosomes [113] and can cause drug resistance [114].

Targets
An inherent difficulty in designing compounds to block the trypanosomatid folate pathway is the need for multi-target design. Several attempts have been made to achieve this with differing levels of success. Another problem in designing anti-trypanosomatid antifolates is the nonstandard physicochemical properties of the target pockets. Pockets of the explored targets such as LmPTR1 are relatively hydrophilic and not the most druggable [41]. This may complicate the ligand design process by, e.g. the need to include structural waters. To add to the complexity, optimizing polar compounds also requires considering interactions with specific transporters [41].
Many of the above problems are compounded by the lack of sufficient biochemical, kinetic, or structural knowledge about the trypanosomatidic drug targets, cross-membrane transport mechanisms and metabolism. For example, substrate inhibition of PTR1 was reported [37,43], which might affect the design of effective PTR1 inhibitors. Mathematical modeling could provide a better understanding of the kinetics of the trypanosomatid folate pathway and therefore a better basis for target validation. However, obtaining reliable kinetic models again requires first filling the gaps in biochemical knowledge for the parasite folate pathway, particularly that of T. cruzi.
The aforementioned lack of structural data on the potential trypanosomatid targets could be partly addressed by tapping the significant advances in predictive power of new protein structure prediction methods, such as AlphaFold [115] and RoseTTAfold [116]. Predicting trypanosomal and leishmanial targets by such approaches has to overcome the poor sequence sampling in databases for these species, and thus the low quality of multiple sequence alignments, but first solutions are appearing, such as the resource made available by Wheeler [117]. Another important question is the suitability of the such predicted structures for SBDD [118,119]. Currently, these methods display problems in capturing pocket conformations, proteinligand complexes, and protein dynamics, e.g. cannot generate an ensemble of active/inactive states for the enzyme [119].

Predictive problems of docking and MD-based approaches
Many of the SBDD studies use docking-based approaches, which tend to predict false positives. The problems with predicting active compounds and ranking them properly may arise from the limitations of the docking approaches, such as lack of correlation between the scores and experimentally measured binding free energies [108]. Compound ranking may be improved by using more rigorous (and more computationally expensive) techniques like free energy perturbation or machine-learningbased free energy estimation methods, like OnionNet-2 [120].
Classical mechanics force fields may not capture the electronic structure of inhibitor candidates accurately enough, for example, the interactions of halogens and the formation of halogen bonds are typically not accounted for [109]. Attempts to overcome those shortcomings by utilizing QM methods have been made [98,99]. Furthermore, entropic contributions to binding free energies are often neglected due to the high computational cost and problems with convergence [121]. The omission of entropy would, for example, affect the pteridine inhibitors, which mostly have long, flexible tails [59], for which the entropic contribution might be particularly important. Another aspect that has not been considered so far in targeting the trypanosomatid folate pathway is ligand binding kinetics, which is regarded as an important factor for drug efficacy [122].
Finally, as mentioned above, the substrates and binding pockets of PTR1s are rather hydrophilic, which means that water bridges and other solvation effects are important for ligand binding (see review by Spyrakis et al. [123]). This requires the inclusion of structural water in docking calculations, with the resulting challenge of selecting the key water molecules to include. They might be predicted based on the alignment of the existing crystal structures [59,66,67,72], based on MD simulations, or from other approaches such as the 3D-RISM method [124]. The water network, however, changes depending on the ligand bound and may affect the reliability of docking results.

Declaration of interest
R C Wade is a patent holder of the patent 'Use of pteridine reductase inhibitors for the prevention and / or treatment of parasitic infections' (https://iris.unimore.it/handle/11380/649260?mode=full). The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures
Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.