Multitarget, selective compound design yields picomolar inhibitors of a kinetoplastid pteridine reductase 1

The optimization of compounds with multiple targets in the drug discovery cycle is a difficult multidimensional problem. Here, we present a systematic, multidisciplinary approach to the development of selective anti-parasitic compounds. Efficient microwave-assisted synthesis of pteridines along with iterations of crystallographic structure determination were used to validate computational docking predictions and support derivation of a structure-activity relationship for multitarget inhibition. This approach yielded compounds showing picomolar inhibition of T. brucei pteridine reductase 1 (PTR1), nanomolar inhibition of L. major PTR1, along with selective submicromolar inhibition of parasitic dihydrofolate reductase (DHFR). Moreover, by combining design for polypharmacology with a property-based on-parasite optimization, we found three compounds that exhibited micromolar EC 50 values against T. brucei brucei , whilst retaining their target inhibition. Our results provide a basis for the further development of pteridine-based compounds and we expect our multitarget approach to be generally applicable to the design and optimization of anti-infective agents. when leaving a single compound out from the data set. The optimization direction indicates whether higher or lower values would putatively lead to improved anti-parasitic effects. assay was performed using the Invitrogen Predictor h ERG fluorescence polarization (FP) assay. A membrane fraction containing h ERG (Predictor h ERG membrane) was used together with a red fluorescent high-affinity ligand of the h ERG channel (Predictor h ERG Tracer Red). Displacement of the latter from h ERG by binding of the test compound can be determined in a FP-based format.


Introduction
The World Health Organization has identified 17 Neglected Tropical Diseases (NTDs) that pose a health burden to over 1.4 billion of people. 1,2 Parasites of the Trypanosomatid family are responsible for two potentially lethal insect-vector borne NTDs: Human African Trypanosomiasis (HAT, sleeping sickness), caused by Trypanosoma brucei, and leishmaniasis, caused by the intracellular parasite Leishmania spp. [3][4][5][6][7] Current therapeutics are limited by toxicity, poor efficacy and parasite resistance, thus underlining the need for new chemotherapies. 8,9 One way to identify new anti-parasitic agents is to apply target-based drug design strategies. [10][11][12] The folate pathway enzyme dihydrofolate reductase (DHFR) is a known anti-cancer, antibacterial and anti-malarial target. [13][14][15][16] It provides reduced folates, which are crucial to biological processes like DNA, protein and amino acid synthesis or one-carbon transfer. 14,17,18 In Trypanosomatids, DHFR inhibition was found to be ineffective due to a metabolic bypass via the biopterin-reducing pteridine reductase 1 (PTR1, Figure 1): When DHFR is inhibited, PTR1, which can also reduce folates, is overexpressed and sustains sufficient metabolite levels to ensure parasite survival. Thus, when targeting the folate pathway in Leishmania, both DHFR and PTR1 need to be considered. [19][20][21] In T. brucei, PTR1 was shown to be a potential antiparasitic target in its own right by RNA interference studies. 22,23 Nonetheless, even nanomolar PTR1 inhibitors have so far shown limited anti-parasitic activity in vitro 24,25 , suggesting that targeting the T. brucei folate pathway may also benefit from the consideration of both PTR1 and DHFR.
Screening a set of folate-related compounds against parasitic folate pathway targets previously led to the identification of compounds 6a (methyl-1-(4-(((2,4-diaminopteridin-6yl)methyl)amino)benzoyl)piperidine-4-carboxylate, herein compound 2) and 6b (methyl-1-(4-(((2,4-diaminopteridin-6-yl) methyl) (methyl) amino) benzoyl) piperidine-4-carboxylate, herein compound 1) as submicromolar inhibitors of Leishmania major PTR1 (LmPTR1) with Ki values of 0.10 μM and 0.04 μM, respectively. 26 2 was additionally a micromolar inhibitor of L. major DHFR (LmDHFR) with a weak selectivity for the parasite enzyme over the human DHFR (hDHFR) (Ki of 4 μM vs. 10 μM). In contrast to the parasite DHFR, which is covalently coupled with thymidylate synthase (TS) in a bifunctional DHFR-TS, the hDHFR off-target is monofunctional and shares only about 30% sequence identity with parasite DHFR domains, indicating potential for further selectivity optimization. [27][28][29] The aim of the current work was to optimize pteridine-based compounds for their inhibition of T. brucei PTR1 (TbPTR1) and TbDHFR, in addition to the Leishmania targets, while ensuring selectivity against the off-target hDHFR. The enzymatic evaluation of reference pteridines reported earlier 26,30 and our published comparative study of trypanosomatid folate pathway proteins 31 supported the design of three series of compounds to explore substituents at three positions on the pteridine structure, and a fourth 'merged' series containing permutations of the substituents in the three series. Docking simulations and three new crystallographic complexes of pteridines with TbPTR1 and a complex with LmPTR1, supported the targetbased design approach and the determination of structure-activity relationships. A systematic analysis of correlations between computed physicochemical molecular descriptors and 3 observed anti-parasitic effects allowed us to prioritize promising compounds for synthesis. In total, 26 new pteridine derivatives were characterized experimentally, most of which showed improved target inhibitory profiles for PTR1 and DHFR of both L. major and T. brucei. Among these, we report the first, to the best of our knowledge, picomolar inhibitors of TbPTR1 and several new low nanomolar inhibitors of LmPTR1, which mostly also show selective micromolar to submicromolar inhibition of the parasite DHFR variants. In vitro evaluations of the effect on T. brucei brucei bloodstream forms revealed three new inhibitors with low micromolar to submicromolar EC50 values against the T. brucei parasite.

Reference compounds inhibit both PTR1 and DHFR and adopt a methotrexate-like binding mode
To systematically assess multitarget inhibition, we measured the inhibition of TbPTR1, TbDHFR, LmPTR1, LmDHFR, and the off-targets hDHFR and hTS, by the folate-related anticancer agent methotrexate (MTX) and 7 pteridine-based reference compounds (1, 2, III-VII, Figure 2AB and Table S1, SI). 26,30,32 Although the seven reference compounds were primarily Figure 1. Overview of the pterin activation in the Trypanosomatidic folate pathway when DHFR is inhibited and PTR1 provides a metabolic bypass. The DHFR domain of the bifunctional DHFR-TS would, under normal conditions, reduce biological folates to tetrahydrofolate (THF). THF is converted to 5,10-methylene THF by the serine hydroxymethyl transferase (SHMT) and this metabolite has a central role in amino acid synthesis, protein biosynthesis and one-carbon transfer. It is also required by the TS domain of DHFR-TS to convert deoxyuridine monophosphate (dUMP) to deoxythymidine monophosphate (dTMP), which is necessary for DNA synthesis. PTR1, which primarily reduces unconjugated pterins like biopterin, takes over folate reduction when DHFR is inhibited, thus acting as a metabolic bypass and an important additional target that needs to be inhibited for shutting down the Trypanosomatidic folate pathway. Both proteins are shown in cartoon representation (DHFR domain of DHFR-TS: purple, PTR1: light pink, single monomer of functional tetramer) with the NADPH/NADP + cofactor in sticks with black carbons and the folate substrate in yellow spheres. In PTR1, in addition, an arginine residue from a neighboring subunit pointing into the active site is shown in magenta sticks.
Previously determined crystal structures show that compounds 1 and 2 share a substrate-like pterin orientation in the complex with LmPTR1. 26 Compound 1, in addition, features a second, so-called inhibitor-like (or MTX-like) orientation, with the bicyclic ring system flipped by 180° and rotated by 30° (Figures 2CD and S1 of the SI). 26 Similar observations have been made in crystallographic complexes of TbPTR1 with small pteridine-based inhibitors. 32 Our crystal structure of the ternary complex of TbPTR1 with NADPH/NADP + and the reference compound 1 (PDB-ID 6rx5, resolution 1.42 Å, experimental details: Tables S2-S3, SI) confirms that the diaminopteridinyl moiety of 1 adopts only the MTX-like orientation (Figure 2E), resembling its MTX-like binding mode in LmPTR1 (Figures 2D and S2, SI). Docking studies consistently indicated that the reference compounds adopt MTX-like binding modes in the different targets and the off-target hDHFR (Table S4, Figure S5, SI). Since the reference compounds showed micromolar to submicromolar inhibition of parasitic PTR1 and DHFR, whilst being modestly selective for parasite over hDHFR, we next aimed to improve the target inhibition profiles in a multitarget-based design approach.

Four series of compounds designed to improve dual target selective inhibition
To optimize the compounds for TbPTR1/TbDHFR and LmPTR1/LmDHFR while minimizing inhibition of hDHFR, we employed our published optimization guidelines for MTX-like scaffolds derived from the extensive comparison of on-and off-targets in the parasitic folate pathway: 31 Based on the overlapping properties in the different protein targets and differences between targets and off-targets, modifications for improved target interactions and selectivity were defined for 1 ( Figure 3A). We designed three series of compounds, each with modifications in one of three parts of the pteridines: the N10 position, the para-amino benzoic acid (PABA) moiety and the tail portion ( Figure 3B).
In the N10 series, five novel pteridines (3-5, 9 and 22, Figure 3B) were modified in the N10 position to improve interactions with PTR1 and parasite DHFR and to exploit the differing pocket sizes and residues in parasitic targets and the hDHFR off-target. 31 Nonpolar substitutions, like the ethyl and propargyl substituents of 3 and 5 were designed to interact with aliphatic residues surrounding N10 in the parasite targets, e.g. Leu209 of TbPTR1; Ile47 and Leu90 of TbDHFR; Leu226 and Leu229 of LmPTR1; Ile20 and Val62 of LmDHFR ( Figure 3A).

6
The large benzyl substituent of 9 may in addition allow for enhanced selectivity due to the lower Based on Panecka-Hofman et al. 31 , this map provides an overview of residues having similar properties in the surrounding of specific ligand portions in all targets (covering only those applied for the design as shown in (B), for full maps, see Figures S3,S4). In some positions, parasite DHFR differs in amino acid type from the offtarget hDHFR, thus highlighting suitable substitution points to improve selectivity. Differing hDHFR residues are labeled in the top right corner of the corresponding parasite DHFR residue. A surface representation of complexes of 1 with TbPTR1 (top right, PDB-ID 6rx5) and TbDHFR (bottom right, MTX-like docking result that was top-ranked by the Glide docking score in PDB-ID 3rg9) highlights the solvent-exposure of the tail in PTR1, which is in contrast well-enclosed in DHFR. All residues in (A) are colored by type: red=acidic, blue=basic, green=polar, yellow=nonpolar. Compound 1 and the NADPH/NADP + cofactor are shown in sticks with cyan and black carbons, respectively. (B) Synthesized members of each designed series with the series for the three modification sites (N10, PABA, tail) shown in a framed box along with the key objectives addressed with the respective modifications. A fourth, merged series, shown at the bottom of the figure, was constructed by combining all modifications studied in silico in all possible permutations. Of the resulting 2014 compounds, six selected representatives were synthesized. The small gray numbers indicate the pteridine ring nomenclature. volume of the hDHFR compared to the parasite DHFR pocket (pocket volume TbDHFR 353 Å 3 , LmDHFR 384 Å 3 and hDHFR 347 Å 3 ). In 22, we replaced N10 by sulfur and the PABA benzene ring by pyridine, to improve off-target selectivity. Whereas hDHFR favors hydrogen bond donors in the proximity of N10 and the PABA ring system, the parasite DHFRs show regions favorable for interaction with hydrogen bond acceptors. 31 Corona et al. 30 demonstrated improved selectivity for PTR1 over hDHFR by hydrophilic N10 substitutions. Our data for reference compound III, with a hydroxyethyl substituent, did not confirm this conclusion ( Figure   2A) and our docking simulations indicated that interaction with a conserved structural water might require an unfavorable conformation of the substituent (Figure S5A, SI). To improve interactions between the substituent and water, we therefore elongated the aliphatic linkage to a hydroxypropyl in 4.
In the PABA series with 5 new pteridine derivatives (11, 21, 25, 27-28, Figure 3B), we explored modifications of the PABA moiety and the amide linkage to the tail portion. To improve selectivity by exploiting the different pocket sizes and contact points of hDHFR and the parasite proteins, in 11, we replaced the PABA phenyl group with benzyl and, in 21, the entire PABA moiety by a meta-aminobenzoic acid (Figure 3). The compound tails are solvent-exposed in PTR1, thus having poorly defined interactions (Figure 3A), whereas strong interactions occur with the hDHFR off-target. 31 We therefore shortened the tail region to achieve full enclosure in the PTR1 binding pocket by replacing PABA by naphthalene or benzene (non-substituted or substituted with -CF3; 25, 27 and 28, respectively).
In the tail series, we explored tail modifications resulting in 10 additional new pteridines (6-7, 10, 14-18, 23 and 26, Figure 3B). In both T. brucei targets, hydrophobic contact points accessible to the compound tail region exist ( Figure 3A). Since the flexibility of the tail likely contributes entropically to binding affinity in PTR1, directional interactions of the tail may be unfavorable. Hydrophobic interactions are geometrically less restrained than, for instance, hydrogen bonds, and thus less likely to entropically penalize binding. Therefore, the methyl ester in the tail of 1 was replaced by the more flexible ethyl and propyl in 7 and 6, respectively.
Additionally, in parasite DHFR, the tail is enclosed by more hydrophobic moieties than in the hDHFR off-target ( Figure 3A) and the surrounding residues show different conformational variability in the crystal structures, which suggests further on/off-target selectivity benefits. 31 Combining the exploitation of these differences with improved enclosure in PTR1 (Figure 3A), we modified the tail to an unsubstituted piperidine (10) or replaced piperidine with an unsubstituted benzene (23). Compound 26, with benzene attached via a flexible ethyl linkage to an MTX-like amide, can benefit from nonpolar and aromatic contact points in PTR1 and DHFR and adapt to their differing placement. The flexible aromatic tail may further form cationπ interactions with positively charged residues in the entrance of the DHFR pocket ( Figure   3A). In compounds 14 and 15, we explored an altered geometry with a one-carbon spacer between N10 and PABA and amide-linked methoxylated tail portions ( Figure 3B). The methoxylations may form additional contacts with hydrophobic residues in the target pocket entrance regions (e.g. Pro99 of TbPTR1, Figure 3A). In addition, an etheryl linkage to a nonsubstituted (18), methoxylated (16) or trimethoxylated (17) benzyl group was explored to likewise exploit hydrophobic, aromatic and positively charged contact points found around the tail region in the various targets ( Figure 3A).
In a fourth series, the 'merged' series, the modifications in the above three series were permuted and merged by decomposing the compounds into fragments and recombining them  (Schemes 2-7). 33 The PABA amine functionalization was achieved by selective alkylation of primary amines to secondary amines using nitriles as an alkylating reagent with Pd/C for intermediates 32-33. 34,35 Conventional alkylation of the latter with propargyl bromide or   The reductive alkylation of amines using nitriles was also used to obtain 51 and 74 in

Designed pteridine derivatives have improved on-target and off-target enzyme inhibitory activity profiles and bind in an MTX-like orientation
The measured inhibitory activities of compounds 3-28 against the targets TbPTR1, TbDHFR, LmPTR1, LmDHFR, and the off-targets hDHFR and hTS, are given in Figure 4 and Table S1.
Most of the new pteridine derivatives displayed 1-2-fold greater inhibition of TbPTR1 than LmPTR1 and were more or equally active against PTR1 than the reference compound 1. The inhibitors with nanomolar to picomolar PTR1 inhibition showed improved selectivity for PTR1 over the off-target hDHFR by up to about 3 orders of magnitude. The activities against TbDHFR  The structural characterization of LmPTR1 in complex with 4 ( Figure 5B) shows the presence of a functional enzyme tetramer in the crystal asymmetric unit with a similar structure to those previously determined. 37,38 The MTX-like binding mode adopted by 4 in LmPTR1 closely resembles that observed in TbPTR1 except for the terminal piperidine moiety ( Figure S2CD of the SI). The latter is highly flexible and was only poorly visible in our models -a possible orientation is reported, but further orientations cannot be excluded.   Table S1. In all panels, interacting residues and the NADPH/NADP + cofactor are shown in sticks (carbons colored according to protein and black, respectively). Hydrogen bonds are represented by brown dashed lines. Docking results are only presented for N1-deprotonated compounds, but similar orientations were observed for the N1protonated variants (see Figure S6). Further IF docking solutions are shown in Figures S7, S8.
The predicted orientations of the compounds in PTR1 were overall similar to the crystallographic data ( Figure 5AB), commonly showing the MTX-like orientation. 9, when docked in presence of conserved structural water, was an exception due to its bulky benzyl substituent and required the reorganization of Trp221 in TbPTR1 (see the induced fit docking pose in Figure 5C). This reorganization is plausible since Trp221 was previously identified as a flexible residue in the TbPTR1 pocket gating region on the basis of crystallographic data. 32 All compounds were roughly similar to 1 in parasite DHFR inhibition (1 IC50 TbDHFR and  The PABA benzene and piperidine of 10 compete for interactions with Phe94 of TbDHFR, which thereby becomes exposed to the solvent. In hDHFR, the corresponding exposed residue is the polar Asn64 and the tail of 10 can interact with Phe31 deeper in the pocket, rendering the mode of binding more favorable in hDHFR. The results are presented for N1-deprotonated compounds, but similar observations were made with N1 protonation (Figure S6).

The activity against T. brucei is related to the hydrophobicity of the compounds
We determined the anti-parasitic effect on T. brucei brucei Lister 427 bloodstream forms and L. infantum intramacrophage amastigotes ( Figure 7B and Table S7, SI). The pteridines were mostly inactive against L. infantum and, despite mostly being nanomolar inhibitors of TbPTR1 and micromolar inhibitors of TbDHFR, many of the new derivatives did not exceed 50% T.
brucei inhibition at 10 μM compound concentration. Consistently, the multiple correlation coefficient between the TbPTR1 and TbDHFR IC50 values and the T. brucei bloodstream form  40 were removed. As the next step, the compounds were filtered by the docking scores obtained for the different targets TbPTR1, LmPTR1, TbDHFR and LmDHFR, followed by evaluation of whether the QikProp properties that were found to correlate with antiparasitic activity (QPLogPo/w: octanol-water partition coefficient; QPlogKp: skin permeability; QPlogKhsa: binding to human serum albumin; coh. index: Index of cohesive interaction in solids, (no. of hydrogen bond acceptors x no. of hydrogen bond donors x 0.5 / surface area) 41 and CIQPlogS: conformation-independent predicted aqueous solubility) were within the range typical for drug-like compounds. Finally, it was evaluated whether the compounds occurred at least twice among the best resulting 30% of compounds for each individual property, as indicated in the bar chart on the right-hand side by the green and blue regions for docking and QikProp results, respectively. All docking results are reported in kcal/mol. The aqueous solubility, S, is reported in mol dm -3 . For target docking, the best docking scores and for off-target docking, the worst docking scores were considered favorable. QikProp results were evaluated based on the correlation with the anti-parasitic activity, with high values for correlating (QPlogPo/w, QPlogKp, QPlogKhsa) and low values for anti-correlating properties (coh. index, CIQPlogS). The bar plot (right) shows the range of the final set of 600 candidate compounds with black dots reporting the individual values for every property. 6 representatives were selected for synthesis in the merged series, such that they span the covered property range, as indicated by the colored diamonds. (B) Percentage inhibition against T. brucei brucei for reference compounds and members of the N10-, PABA and tail-modified series (left) and the selected representatives of the in silico merged library (right). The average of at least three independent determinations is shown with the standard deviation. The inactive compounds in the tail modified series, 14, 17 and 18 were omitted. Activities can be found in Table S7

20
We considered the correlating predicted properties as an additional prioritization filter for the in silico merged library, see Figure 7A. Of the six synthesized compounds, 19, 20 and 24 showed an improved percentage of T. brucei inhibition at 10 μM, as was suggested by their properties. For those compounds, EC50 values were determined, see Table 2. Indeed, the three more lipophilic compounds were found to have low micromolar EC50s against T. brucei brucei with 24 being the best (EC50 0.66 ± 0.48 μM) and they showed SIs of 3-38 on the basis of their cytotoxicity on THP-1 derived macrophages.

Conclusion
Applying a multitarget-based approach to the development of novel therapies for HAT and leishmaniasis, we here focused on pteridine-based inhibitors of L. major and T. brucei PTR1 and DHFR and successfully designed the first known picomolar inhibitors of TbPTR1. While LmPTR1/LmDHFR inhibition was previously explored for this compound class, we here demonstrated the potential of pteridine-based inhibitors against the TbPTR1/TbDHFR system. 27 We solved a crystal structure of the reference compound 1 bound to TbPTR1 to confirm the overlap in observed binding modes between the two PTR1 variants and the   Tables S8,S9. of crystal structures of complexes and computational docking enabled us to obtain a complete characterization of the binding modes of the pteridines to their molecular targets and supported the derivation of a SAR. The compounds were also tested against the human off-targets hDHFR and hTS. While they were sometimes only modestly selective for the parasitic DHFR variant, many showed 1000-fold and higher selectivities for PTR1 over the off-targets and thus, the novel PTR1 inhibitors can overall be considered selective for the parasite proteins.
While many compounds exhibited excellent inhibitory activity at the target level, they were often only modest inhibitors of T. brucei brucei bloodstream forms and inactive towards L.
infantum intracellular amastigotes in vitro. We found that increased lipophilicity correlated with improved effects on T. brucei. We were able to prioritize compounds from an in silico library for synthesis by using predicted ADMET-related properties which suggested a likely improvement of the trypanocidal effect. In this way, we found three improved compounds,19,

24
The precipitate was then collected by filtration and dried before the final compound was purified by fractional crystallization from methanol, DCM and Et2O. Crystals of LmPTR1 were prepared as described elsewhere. 38 The LmPTR1-cofactor-crystallization drop. After 5 h, crystals were transferred to the cryoprotectant solution and flash frozen in liquid nitrogen.

General
Data collection, structure solution and refinement. X-ray crystallographic data were collected using synchrotron radiation at the Diamond Light Source (DLS, Didcot, United Kingdom) beamlines I04-1 and I03 equipped with a Dectris Pilatus 6M-F and a Pilatus3 6M detector, respectively. Reflections were integrated using MOSFLM and scaled with Scala (CCP4 suite). [45][46][47][48][49] Data collection and processing statistics are reported in Table   S2. The crystals of TbPTR1 and LmPTR1 belonged to the primitive monoclinic space group P21 and the primitive orthorhombic space group P212121, respectively. Both had a functional enzyme tetramer in the asymmetric unit. The structures were solved by molecular replacement using MOLREP and either a TbPTR1 (PDB-ID 5jdc) or a LmPTR1 tetramer (PDB-ID 5l4n) as the searching model (all non-protein atoms were excluded). 38,42,50 Models were refined using REFMAC5 (CCP4 suite). 51 Visual inspection and manual rebuilding of missing atoms was performed using Coot. 52,53 Water molecules were added with the automated standard procedures implemented in the software ARP/wARP and checked with Coot. 54 In the higher resolution complexes  Table S3. Figures were generated using CCP4mg. 56 Coordinates and structure factors were deposited in the Protein Data Bank under the PDB-IDs 6rx5 (TbPTR1-NADPH/NADP + -1), 6rx0 (TbPTR1-NADPH/NADP + -3), 6rx6 (TbPTR1-NADPH/NADP + -4), and 6rxc (LmPTR1-NADPH/NADP + -4).

TbPTR1, TbDHFR, LmPTR1, LmDHFR, hDHFR and hTS target/off-target enzyme assays.
In vitro assays for TbPTR1 and LmPTR1 were based on the coupled assay reported by Shanks et al. 57 The assay non-enzymatically links the reduction of cytochrome c (Cc) with the reduction of dihydrobiopterin to tetrahydrobiopterin, catalyzed by PTR1. The formation of reduced Cc (Fe 2+ ) results in a signal increase in the photometric readout at 550 nm wavelength. TbPTR1 and LmPTR1 assays were performed in a buffer containing 20 nM sodium citrate (pH 6.0) in a well-plate-based format as previously reported. 42 LmDHFR, TbDHFR, hDHFR and hTS activities were assessed spectrophotometrically according to published procedures. 58,59 Each inhibitory compound was assayed at five different concentrations in duplicate (confidence interval ρ < 0.05) and IC50 values were calculated as described in the SI. All receptors were prepared in the presence of MTX (from the following PDB-IDs for The validation of the chosen docking protocol is presented in the SI.

Computational property prediction, Pan-assay interference compounds (PAINS) and
correlation analysis with anti-parasitic data. Physico-chemical descriptors and parameters related to ADMET were computed for all prepared compounds using QikProp (Schrödinger). 39 We then used Python scripts and SciPy to compute Pearson correlations (R), R 2 values and two-tailed P-values for each property with the measured percentage of inhibition of T. brucei at 10 µM compound concentration. Only properties with a statistical significance level a no more than 0.05 were considered further. To ensure robustness against leaving single compounds out of the analysis, we performed an additional resampling analysis by leaving every compound out once before recomputing the correlations. Properties with R above 0.40 or below -0.40, P-value £ 0.05 and being identified in more than 50% of the resampling correlation analyses were considered to be the most robust markers for the optimization for anti-parasitic effect. These properties were employed to prioritize compounds for synthesis as part of series 4 as demonstrated in Figure 7 and explained in detail in the SI.
In addition, a multivariate correlation coefficient between parasite target protein inhibition and anti-parasitic activity was determined. Details on the correlation calculations are reported in the SI.
Finally, all synthesized compounds were checked for PAINS (PAINS filters A, B and C), undesirable substructure moieties, covalent inhibition, and compliance with the rule-offive with the FAF-Drugs4 webserver (fafdrugs4.mti.univ-paris-diderot.fr/) by inputing SMILES strings for the compounds. 77 In vitro biological evaluation against T. brucei and L. infantum intramacrophage amastigotes. The efficacy against T. brucei brucei Lister 427 bloodstream forms was evaluated in a modified resazurin-based assay as previously described. 78 Cells were grown at 37°C and 5% CO2 in a complete HMI-9 medium supplemented with 10% fetal calf serum (FCS) and 100 UI/mL of penicillin/streptomycin. Cultures were then diluted to a cell density of Cytochrome P450 (CYP450) assays against isoforms 1A2, 2C9, 2C19, 2D6 and 3A4 were performed using the Promega P450-Glo assay platform. Microsomal preparations of cytochrome P450s from baculovirus-infected insect cells were used. In this assay, light is generated when a CYP450 enzyme acts on its substrate and a decrease thereof was indicative of inhibitory effects of the tested compound on the respective isoform. 42 For monitoring mitochondrial toxicity caused by the test compounds in the 786-O cell line, uptake of MitoTracker Red (chloromethyl-X-rosamine) combined with high content imaging was used. Cells were maintained in Rosswell Park Memorial Institute (RPMI)-1640 medium containing 2 mM glutamine, FCS (10% v/v), streptomycin (100 µg/mL), and penicillin G (100 U/mL). 42 The cytotoxicity assay against A549 cells was performed using the CellTiter-Glo assay from Promega. The number of viable cells present is directly proportional to the cellular ATP content, which is detected. The A549 cells were obtained from DSMZ (German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany) and grown in Dulbecco's modified Eagle medium (DMEM) with FCS (10% v/v), streptomycin (100 µg/mL) and penicillin G (100 U/mL). 42

Notes
The authors declare no competing interests.