Isomerization of antimalarial drug WR99210 explains its inactivity in a commercial stock

WR99210, a former antimalarial drug candidate now widely used for the selection of Plasmodium transfectants, selectively targets the parasite dihydrofolate reductase thymidine synthase bifunctional enzyme (DHFR-TS) but not human DHFR, which is not fused with TS. Accordingly, WR99210 and plasmids expressing human dhfr have become valued tools for the genetic modification of parasites in the laboratory. Concerns over the ineffectiveness of WR99210 from some sources encouraged us to investigate the biological and chemical differences of supplies from two different companies (compounds 1 and 2). Compound 1 proved effective at low nanomolar concentrations against Plasmodium falciparum parasites, whereas compound 2 was ineffective even at micromolar concentrations. Intact and fragmented mass spectra indicated identical molecular formulae of the unprotonated (free base) structures of 1 and 2; however, the compounds displayed differences by thin layer chromatography, reverse phase high performance liquid chromatography, and ultraviolet-visible spectroscopy, indicating important isomeric differences. Structural evaluations by 1H, 13C, and 15N nuclear magnetic resonance spectroscopy confirmed 1 as WR99210 and 2 as an isomeric dihydrotriazine. Induced fit, computational docking models showed that 1 binds tightly and specifically in the P. falciparum DHFR active site whereas 2 fits poorly to the active site in loose and varied orientations. Stocks and concentrates of WR99210 should be monitored for the presence of isomer 2, particularly when they are not supplied as the hydrochloride salt or are exposed to basic conditions that can promote isomerization. Absorption spectroscopy may serve for assays of the unrearranged and rearranged triazines.


INTRODUCTION
WR99210 [4,6-diamino-1,2-dihydro-2,2-dimethyl-1-(2,4,5-trichlorophenoxypropyloxy)-1,3,5 triazine] is a folate pathway antagonist with potent activity against Plasmodium malaria parasites (1). Action of WR99210, originally designated as BRL 6231, in parasites is against the bifunctional dihydrofolate reductase thymidylate synthase enzyme (DHFR-TS), where the compound binds to the DHFR active site and blocks the production of tetrahydrofolate in the folate pathway (2). In contrast, WR99210 interacts only weakly with human DHFR (hDHFR). Episomal expression of the hDHFR coding sequence in Plasmodium falciparum produces a 4-log increase in WR99210 halfmaximal effective concentration (EC50) against transformed parasites, providing a powerful selectable resistance marker for transfection studies (3). The exciting potential of WR99210 as an antimalarial drug candidate eventually waned after preclinical and clinical trials demonstrated adverse events including severe gastrointestinal side effects in non-human primates and humans, even at low drug doses (4)(5)(6)(7). Pursuit of a pro-drug form of WR99210 (PS-15) was also limited by a regulatory restriction on the starting material, 2,4,5 trichlorophenol, used to produce dioxin, among other toxic substances (8,9). However, alternative antifolate compounds that incorporate a flexible linker as seen in WR99210 are under evaluation, including candidate P218 which is in clinical trials with the Medicines for Malaria Venture (10)(11)(12)(13).
Genetic transfections and transformations of Plasmodium have become elemental tools of malaria research. Positive selection of hDHFR-transformed parasites is commonly achieved by WR99210 exposure in vitro and in vivo with P. falciparum as well as certain other Plasmodium species (14)(15)(16). Untransformed Plasmodium spp. are sensitive to nanomolar levels of WR99210, and spontaneous development of WR99210 resistance has not been reported from exposure to the compound (17). The usefulness of the compound for the selection of hDHFR-transformed lines thus created an ongoing demand for WR99210 for genetic manipulations of the parasites.
In recent transfection experiments, we observed that WR99210 from one source (stock JP, from Jacobus Pharmaceutical Company Inc., Princeton, NJ ) was effective at nanomolar concentrations, but a stock of WR99210 from another source (stock SA, from Sigma-Aldrich Corp. -St Louis, USA) did not kill untransformed P. falciparum parasites, even at micromolar concentrations. After this discovery, and learning of similar observations reported in online scientific discussion groups (18,19), we initiated investigations into potential differences between the JP and SA WR99210 stock compounds. Here we report results from drug response assays, chemical and structural evaluations, and computational modeling studies that demonstrate dramatic effects on activity from isomeric differences between the JP and SA stocks. Further, we propose a mechanism by which the inactive isomer may develop from the active WR99210 compound, and we suggest some methods for rapid detection of the inactive isomer.

P. falciparum parasites exhibit strikingly different responses to stocks of WR99210 from two sources
For more than 20 years (3), our standard EC50 assays with JP WR99210 have routinely yielded susceptibilities in the sub-nanomolar range for untransformed P. falciparum parasites. Results for two P. falciparum lines are listed in Table 1. NF54 encodes an antifolate sensitive isoform of PfDHFR (N51, C59, S108), while Dd2 (N51I, C59R, S108N) encodes mutations at PfDHFR residues that confer resistance to certain other antifolates such as pyrimethamine, but not to WR99210. NF54 EC50 results with JP WR99210 were 0.056 nM with a 95% confidence interval (CI) of 0.029 -0.103 nM, while Dd2 showed a 10´ increase to 0.62 nM (CI: 0.580 -0.671 nM). In striking contrast to these results with JP WR99210, our EC50 results from stocks of more recently acquired SA WR99210 (Sigma catalog no. 47326-86-3 lots/batches 0000042122, 0000014239, 095M4622V, 070M4610V) were 900 -23,000 fold higher: 1.25 µM (CI: 1.01 -1.57 µM) for NF54 and 547 nM (CI: 525 -571 nM) for Dd2 (Table 1). Morphological appearances of parasites exposed to stock compounds of JP WR99210 and SA WR99210 were consistent with results of the EC50 assays. Culture samples of synchronized NF54 schizont-infected erythroctyes were exposed to 2 nM WR99210 from JP or SA, to 1 µM WR99210 from SA, or to 0.0002% DMSO (control) for 4 days.
Microscopy of Giemsa-stained thin smears of these exposed cells showed dead and dying parasites only in response to 2 nM JP WR99210; the appearances of the parasitized cells under all of the other exposures remained healthy, including parasites exposed to 1 µM SA WR99210.

An isomeric difference is suggested by TLC, RP-HPLC, and UV-vis spectroscopy
We next checked for evidence of a structural difference between 1 and 2 that could explain their different effects in the drug response assays. Thin-layer chromotography (TLC), reverse phase high performance liquid chromatography (RP-HPLC), and ultraviolet-visible spectroscopy (UV-vis) were performed with the samples dissolved in tetrahydrofuran:methanol (2:1), which provided solublities of more than 3 mg/mL vs. 0.2 mg/ml with DMSO. Silica gel TLC of these samples developed with a chloroform:diethyl ether:methanol solvent system showed different retention factors (Rf) for 1 (Rf = 0.30) and 2 (Rf = 0.37) (Fig. S3). RP-HPLC experiments indicated that both compounds were more than 95% pure and eluted with different retention times (tR) of 13.7 minutes for 1 and 13.1 minutes for 2 ( Fig. S4-5).
Consistent with the differences between 1 and 2 detected by TLC and RP-HPLC, UVvis spectroscopy showed the presence of two separate compounds. These differences were particularly evident in the different absorbances in the plateau region between 230 -240 nm (Fig. 1B). The results of these analyses and the drug assays pointed to the likelihood that 2 was an inactive isomer of the WR99210 structure 1.

Chemical derivatization and nuclear magnetic resonance spectroscopy clarify structural differences between compounds 1 and 2
We next sought to determine the structures of 1 and 2 by 1-and 2-dimensional 1 H, 13 (Table S1, Fig. S6-13). Due to the lack of protons in the triazine ring and the four-bond separation between C-9 and the closest carbon, we were unable to unambiguously assign the chemical shifts for the triazine ring from 1 H-13 C data alone. We therefore recorded a 1 H-15 N HMBC ( 2,3 JHN) experiment to assign the NH and NH2 chemical shifts and the regiochemistry of the ring. The structure of compound 1 was therefore as published for WR99210 (20,21). NMR spectroscopy of compound 2 showed that the 1 H and 13 C NMR data corresponding to the (2,4,5-trichlorophenoxy) propoxy moiety, extending from C-1 through C-9, were in close agreement with 1. However, the chemical shifts for the diamino-dimethyl triazine unit were absent, and the 1 H spectrum showed broad peaks corresponding to two NH groups at dH 5.86 and 6.80 and one NH2 group at dH 5.61.
Together, both the MS data and the NMR data suggested we had observed two, distinct singly-charged species by HRMS; namely a protonated and positively charged compound 2 or [2+H] + , and a positively charged compound 1, or [1] + already present in its protonated form (Fig. S2). The NMR data therefore demonstrate that 2 is an isomer of compound 1. We recorded 1 H-15 N HSQC and HMBC spectra for 2 at multiple temperatures and in different solvents; however, unlike the case for compound 1, we were unable to determine the complete structure of 2 by NMR alone (Table S2,  To obtain additional NMR data we permethylated 2 using a stoichiometric excess of methyl iodide in sodium carbonate buffer (Fig. 2B). HRMS of the product indicated a molecular formula of C19H29Cl3N5O2, which corresponded to the addition of five methyl groups (Fig. S19). These were confirmed by signals from the 1 H and 13 C NMR spectra ( Fig. S20-21, Table S3). The chemical shifts of all 1 H, 13 (Fig S22-25).

Modeling of compounds 1 and 2 within the PfDHFR-TS binding pocket
Using the induced-fit docking mode of Glide, we examinined the predicted binding of Results confirmed that our model was in good agreement with the known binding geometry (Table S4) (Fig. 3B). The bestscoring docking pose with 1 had a GlideScore of -9.33 and an induced fit score of -425.81, while the best-scoring pose of 2 produced a GlideScore of -8.07 and induced fit score of -423.44 (Fig. 3C). Furthermore, the best scoring binding pose of 2 was in an orientation opposite to that of 1 and the crystallographic pose, with the halogenated ring placed into the interior the enzyme rather than protruding from the pocket (Fig. 3A).

DISCUSSION
Although not suitable for clinical use, WR99210 continues to be an important tool in molecular parasitology as a selection agent for genetically modified parasites. Recent observations of greatly different efficacies of WR99210 stocks from different sources have been unexplained. Here, we show that isomerization of WR99210 accounts for these efficacy differences, and we propose a rearrangement mechanism for isomerization. The potent antifolate activity of WR99210 that results from tight and specific binding to the Plasmodium DHFR-TS site is lost with the inactive isomer, which interacts only weakly at the DHFR active site in various, greatly different orientations.
HRMS analysis of WR99210 (compound 1) and its isomer (compound 2) showed that molecular formulae of these compounds were identical, without evidence in the stocks for degenerative products of reduced size or substantial impurities, incidental polymerization, or redox changes. HRMS also confirmed the experimental concentrations of 1 and 2 in DMSO-dissolved stocks used for drug response assays, thus demonstrating that the expected exposures of these compounds to parasitized erythroctyes were the same. Having eliminated these possible explanations for the large, supplier-dependent efficacy differences between the stocks, we performed TLC, RP-HPLC, and UV-vis spectroscopy studies on stocks of 1 and 2 for evidence of structural differences. All three methods identified differences between 1 and 2. demonstrating that it is a positional isomer of 1, is identical to that of a previous report, which demonstrated production of 2 from 1 in an ethanolic solution brought to pH >8 with sodium hydroxide, or with added triethylamine (21). Conversion of arylmethoxydimethyl-dihydrotriazines from their base form to isomeric dihydrotriazines had also been observed when the compounds were heated in ethanol, benzene, or partial aqueous suspension (24). Stocks of WR99210 may thus be inactivated by spontaneous rearrangement to isomer 2 under basic conditions. Isomer 2 differs from WR99210 by a repositioning of the propoxyl substituent on the triazine ring. Figure 4 presents a proposed pathway of isomerization by a basemediated ring opening, followed by ring closure at the gem dimethyl carbon with the NH2 group proximal to the propoxy linker. First, WR99210 forms from a substituted biguanide by a pathway involving O-bonded amine attack at the dimethyl-bearing methine carbon to form the triazine ring (8,24,25). Second, the amine groups of the diaminotriazine moiety in WR99210 are stabilized by the HCl salt form. When converted to, or purified as, the free base, tautomerization of the guanidine groups can occur, allowing for the prior imine nitrogen (now an NH2) to attack the dimethyl-bearing methine carbon to form the dihydrotriazine isomer 2. Figure 4. Proposed mechanism for the isomerization of compound 1 to 2. The mechanism of isomerization relies upon the conditions for production of the freebase rather than the hydrochloride salt, in agreement with the solved structures (Fig 2). The proposed mechanism starts with base-mediated ring opening, followed by ring closing via substitution at the gem dimethyl quaternary carbon by the NH2 group proximal to the propoxy linker. The isomerization which repositions the amine substituents and extends the propoxy linker between the halogenated ring and the triazine ring.
The results of our docking modeling indicate that isomer 2 binds to PfDHFR-TS with much lower affinity than WR99210, if at all. This dramatic difference of affinity further verifies the Plasmodium DHFR active site as WR99210's target of action and accounts for the compound's almost complete loss of drug efficacy after isomerization. Stocks and concentrates of WR99210 should therefore be checked for the presence of isomer 2, particularly when they are not obtained and maintained as the hydrochloride salt or are exposed to basic conditions. Among the checks reported here are analysis by UVvis spectroscopy, determination of the RP-HPLC elution time, and verification of the TLC band and its retention factor. We also note that infrared spectroscopy in the 6.0 -10.0 µm region has been reported to distinguish diagnostic differences between unrearranged and rearranged triazines (24). These assessments alone or in combination may serve for accessible quality control of WR99210.

Dose-response assays
In vitro drug response assessments were performed employing a standard 72-hour malaria SYBR Green I assay against the lab-adapted lines Dd2 and NF54 (27)(28)(29). Twofold serial dilutions of JP and SA WR99210 (50 µL) were added across a 96-well plate, reserving two wells per row as drug-free controls. After reaching 4 -10% parasitemia with >70% ring stage parasites, cultures were resuspended to 1% parasitemia and 1% hematocrit in CM and 150 µL was added to each well for drug phenotype response.
EC50 values were determined using the variable sigmoidal function feature from Prism 7 on four independent replicates (GraphPad Software Inc).

HPLC with UV-vis spectroscopy
A SB-C3, 3.5 micron, 300Å, 0.3 ´ 100 mm capillary column (Aligent) was used for chromatography at 6 ml/min. Typical sample amounts were 100 nanograms (ng). The column was equilibrated in 80% of 0.1% formic acid:20% acetonitrile and eluted over 15 minutes with a gradient to 100% acetonitrile. Detection was done at 290 nm with diode array spectral acquisition between 220 and 400 nm. Spectral analysis was performed with the Chemstation 2D software. For HPLC prior to Mass Spectrometry (further methods below) all chromatography conditions were maintained, except, a flow rate of 10 ml/min was used and approximately 25 ng of solubilized sample (primary stock solution) was injected.

Mass Spectrometry
Acquisition and analysis was done with a Sciex 4000 QTrap in positive mode using either HPLC or direct infusion at 1 µg/ml of WR99210 in 85% acetonitrile:15% formic acid (0.1%) with a flow rate of 12 µl/min. Spray voltage was 4000, de-clustering potential was 25 volts and nebulizing nitrogen gas was 20 psi. MS2 fragmentation was done with a collision energy of 40 and MS3 with AFC values of 40 to 65.

Nuclear Magnetic Resonance
NMR spectra were recorded on Bruker Avance 500 and 600 MHz spectrometers equipped with z-shielded gradient cryoprobes. Spectra were recorded at multiple temperatures, and 1-and 2-dimensional homonuclear and heteronuclear 1 H, 13 C and 15 N spectra were recorded. See Supporting Information for additional details.

Computational docking studies
The three-dimensional structure of PfDHFR-TS (PDB ID 1J3I) was downloaded from the Protein Data Bank. The protein structure was readied for docking via the Protein Preparation procedure, and used Induced Fit Docking protocol 2015-2, Glide version 6.4, Prime version 3.7, Schrödinger, LLC, New York, NY, 2015, release 2018-4.
To achieve unbiased docking, the conformation of WR99210 from the PDB structure was not used. Instead, the 3D conformer of WR99210 was downloaded from PubChem using CID 121750. The SA isomer was drawn in the PubChem sketcher and converted to 3D using Open Babel 2.3.1 (30). Both ligands were readied for docking using the LigPrep procedure in Maestro (31).
Induced fit docking (IFD) was performed using the Maestro suite. IFD used Glide for ligand docking with a softened potential to increase the possible initial protein conformations by decreasing the van der Waals repulsion term, permitting closer packing, and a final re-docking after protein optimization. Prime was used to predict side-chain position within each protein-ligand complex and to find stable coformations around the docked ligand. To evaluate the quality of the docked poses, IFD provided an energy score for each docked pose that combines both Glide and Prime. Root mean square deviation (RMSD) calculations were performed between the crystallographic structure and the poses resulting from IFD using the DockRMSD program (https://zhanglab.ccmb.med.umich.edu/DockRMSD/). 2D ligand interaction diagrams of the active site were created within Maestro and virtual reality visualization was done via UCSF ChimeraX (https://www.rbvi.ucsf.edu/chimerax/) and the HTC Vive Pro.