The Key Glycolytic Enzyme Phosphofructokinase Is Involved in Resistance to Antiplasmodial Glycosides

Malaria, caused by Plasmodium parasites, continues to be a devastating global health issue, causing 405,000 deaths and 228 million cases in 2018. Understanding key metabolic processes in malaria parasites is critical to the development of new drugs to combat this major infectious disease. The Plasmodium glycolytic pathway is essential to the malaria parasite, providing energy for growth and replication and supplying important biomolecules for other essential Plasmodium anabolic pathways. Despite this overreliance on glycolysis, no current drugs target glycolysis, and there is a paucity of information on critical glycolysis targets. Our work addresses this unmet need, providing new mechanistic insights into this key pathway.

In recent work, PfPFK9 mutations have been associated with the reversal of fosmidomycin resistance in P. falciparum (7). Fosmidomycin, an antibiotic and antimalarial drug candidate, acts as a competitive inhibitor of a key enzyme in methylerythritol phosphate (MEP) pathway isoprenoid biosynthesis in the apicoplast of the malaria parasite (8). Fosmidomycin resistance has been linked to the functional loss of haloacid dehalogenase 1 (PfHAD1), a sugar phosphatase (9). In fosmidomycin-resistant parasites, loss of PfHAD1 function results in dysregulation of glycolysis and increased flux of triose-phosphates into the MEP pathway, with a concomitant reduction in the effectiveness of fosmidomycin (9). A second HAD family member, PfHAD2, has also been implicated in fosmidomycin resistance (7). Curiously, fosmidomycin sensitivity is restored in PfHAD2 mutants that have been subjected to prolonged culture in the absence of drug pressure. These PfHAD2 mutants gain mutations in PfPFK9 that result in the restoration of fosmidomycin sensitivity (7).
In this study, we provide the first report that PfPFK9 mutations are associated with resistance to an antiplasmodial compound. Previously, we screened a panel of primary sulfonamide glycosides for in vitro growth inhibitory activity against asexual intraerythrocytic stages of drug-sensitive (3D7) and multidrug-resistant (Dd2) P. falciparum parasites. The most potent sulfonamide glycoside (PS-3) exhibited a 50% inhibitory concentration (IC 50 ) value of ϳ1 M and Ͼ40-fold selectivity for P. falciparum versus that for mammalian cells (Fig. 1) (10). In this study, P. falciparum parasites were selected for in vitro resistance to PS-3 (10), followed by whole-genome sequence analysis of clones to identify putative PS-3 targets and/or resistance mechanisms. A point mutation in the PfPFK9 gene was shown to be linked to PS-3 resistance and subsequently confirmed via reverse genetics. While PS-3 did not significantly inhibit recombinant Plasmodium PFK activity, PS-3-resistant P. falciparum parasites with PfPFK9 gene mutations redirected glucose flux into the pentose phosphate pathway, at the expense of upper glycolysis, while still maintaining lower glycolysis and continued ATP production. While the precise target of PS-3 remains undefined, these data suggest that the [20 M]) for Ͼ8 weeks did not result in a significant alteration in the IC 50 (P Ͼ 0.05) compared to that for 3D7-C3 PS3 selected with 10ϫ PS-3 (see Fig. S2). A comparison of in vitro IC 50 values for 3D7-C3 PS3 versus those for 3D7-C3 wild-type parasites showed generally no significant difference for the clinically used antimalarial drugs chloroquine, pyrimethamine, cycloguanil, artesunate, atovaquone, and quinine (P Ͼ 0.05) ( Table 1). Likewise, there was no significant difference in IC 50 values for the clinical candidates KAE609 (11) (PfATP4 inhibitor) and DSM161 (12) (an analogue of DSM265, a dihydroorotate reductase [DHOD] inhibitor [13]). This is reflected in the calculated resistance indices (Ri), which range from 0.8 to 1.6. In contrast, a Ri of 8.3 was obtained when comparing P. falciparum 3D7-C3 and 3D7-C3 PS3 PS-3 IC 50 values (Table 1).
Structure-activity relationship analysis of the P. falciparum PS-3 resistance phenotype. To define which PS-3 structural group(s) contributes to the 3D7-C3 PS3 resistance phenotype, PS-3 analogues were tested in in vitro growth inhibition assays against 3D7-C3 and 3D7-C3 PS3 parasites. Compounds included PS-3=, which lacks the primary sulfonamide (PS) moiety, and four compounds that retain the PS moiety (PS-1, PS-3, PS-7, and PS-10) ( Table 2). All compounds contain a per-O-acetylated glucose type sugar with an acetyl R group and a triazole substituent (Y ϭ I or H) but vary with respect to their glycosidic linkage (X ϭ S, O, or SO 2 ) ( Table 2). Compared to PS-3, the absence of the PS moiety in PS-3= resulted in a significant reduction in activity for both 3D7-C3 PS3 and 3D7-C3 wild-type parasites (ϳ4-fold higher IC 50 s; P Ͻ 0.05) ( Table 2); however, the PS-3= Ri remained similar (Ri, ϳ6) ( Table 2) to that for the PS-3 selection compound, indicating that the PS-3 resistance phenotype is independent of the PS group. Likewise, PS-7 and PS-10 showed a reduction in overall activity against both 3D7-C3 PS3 and wild-type 3D7-C3 but retained similar Ri values to that for PS-3 (Ri ϳ6 to 7) ( Table 2), indicating that the variable glycosidic linkage (X ϭ S, O, or SO 2 ) ( Table 2) does not impact resistance to PS-3. PS-1, which is identical to PS-3 (Table 2) (Y ϭ I) except for the triazole substituent (Table 2) (Y ϭ H) showed Ͼ2.3-fold reduced activity against 3D7-C3 PS3 versus that against 3D7-C3, suggesting that the iodo substituent (Y ϭ I) of the triazole moiety is likely also not contributing to the resistance phenotype of 3D7-C3 PS3 parasites. Overall, these structure-activity relationship (SAR) data implicate the glucose moiety in PS-3 as being associated with the resistance phenotype of 3D7-C3 PS3 parasites.
To determine if sugars other than glucose play a role in the 3D7-C3 PS3 resistance phenotype, PS-3 analogues differing only with respect to the sugar group were tested against 3D7-C3 PS3 and 3D7-C3. 3D7-C3 PS3 was found to be cross resistant to PS-11 (galactose replacing glucose) and PS-12 (glucuronic acid replacing glucose), as shown by similar Ri values to that for PS-3 (Table 2). To a lesser extent, 3D7-C3 PS3 was found to be cross resistant to PS-14 (the disaccharide maltose replacing glucose; Ri, 2.5)  (Table 2). Together, these SAR data suggest that the mode of action of PS analogues containing the sugars glucose, galactose, glucuronic acid, and, to a lesser extent, maltose may be affected by the phenotypic change associated with the 3D7-C3 PS3 resistance. These sugars have subtle structural and stereochemical differences. Interestingly, 3D7-PfHK ϩ , a parasite line that was previously shown to be resistant to the glucose analogue 2-deoxyglucose (14), is not resistant to PS-3 (data not shown).
Whole-genome sequencing of 3D7-C3 PS3 subclones identifies two independent single nucleotide polymorphisms in the P. falciparum phosphofructokinase (Pf-PFK) gene. Three independent 3D7-C3 PS3 subclones were generated (termed 3D7-C3 PS3-1 , 3D7-C3 PS3-2 , and 3D7-C3 PS3-3 ), and their PS-3 resistance profiles were confirmed (Fig. 2C). The IC 50 values of all the clones were significantly different from that of the wild-type 3D7-C3 parasites (Fig. 2C) (P Ͻ 0.05) but not for the control drug chloroquine (P Ͼ 0.05) (Fig. 2D). PS-3 resistance indices of 15.0, 6.6, and 10.8 were observed for 3D7-C3 PS3-1 , 3D7-C3 PS3-2 , and 3D7-C3 PS3-3 , respectively. A comparison of the genome sequences of the three clones versus that for 3D7-C3 revealed 12 new coding mutations, which included two point mutations (L513Q and R860G) in the PfPFK9 gene (PF3D7_0915400) (Fig. 3; Table S1). The L513Q mutation was found only in clone 3D7-C3 PS3-2 and maps to the ␤-domain of PfPK9 (amino acids 1 to 660) (Fig. 3). The R860G mutation was found in clones 3D7-C3 PS3-1 and 3D7-C3 PS3-3 and maps to the ␣-domain (amino acids 777 to 1418) (Fig. 3). Mutations were observed in the P. falciparum multidrug-resistant protein 2 (PfMDR2; Pf3D7_1447900) and the P. falciparum sodium/hydrogen exchanger (PfNHE; Pf3D7_1303500) for the 3D7-C3 PS3-1 and 3D7-C3 PS3-3 clones (Table S1). The likelihood of finding three independent missense mutations in PfPK9 by chance is very low (P ϭ 1.7eϪ10, hypergeometric mean function). Furthermore, in contrast to that for PfMDR2, no PfPK9 mutations were detected in a systematic analysis of the genome sequences of 262 evolved parasites resistant to 37 diverse compounds (15) performed with the same sequence analysis pipeline. No other common gene amplifications were observed in the three independent clones analyzed. Modeling of glycoside resistance alleles. To inform the possible structural basis of compound resistance, we evaluated the effects of resistance alleles (R860G and L513Q) on three-dimensional models of PfPFK9 ␣ and ␤ domains (Fig. 3). Currently, no structure exists for the distinct "long" PFKs observed in apicomplexans. Our structural model shows that the R860G and L513Q mutations do not share cognate regions of their respective domains. The R860G mutation is adjacent to the substrate-binding pocket of the C-terminal ␣ domain, while the L513Q mutation is buried between an alpha-helix and beta-strand of the N-terminal ␤ domain and is not predicted to impact substratebinding or catalytic residues. This suggests that these mutations may share their resistance phenotype through a more general mechanism of overall reduced PFK function rather than a specific change in a particular inhibitor-protein binding interface.
PfPFK9 R860G mutation validated as a resistance mechanism of PS-3. CRISPR/ Cas9-based genome editing was used to validate the contribution of the two PfPFK9 mutations, L513Q and R860G, to the resistance phenotype. Donor templates were synthesized and cloned into a CRISPR vector encoding Cas9 and one of two single guide RNAs (sgRNAs) targeting the relevant locus (see Fig. S3). For each mutation, two donor templates were generated encoding either the putative resistance mutation (e.g., L513Q) or a silent control (e.g., L513L). Additional silent mutations common to both sets of donors were also included to prevent sgRNA binding to the repair locus (Fig. S3). The mutations were introduced into the Dd2 strain, and clonal lines were derived. The R860G-edited line demonstrated a 2.1-fold increase in IC 50 values compared to that for the R860 silent control (P Ͻ 0.0001) (Fig. 2E). There was no significant change in IC 50 values for the L513Q-edited line compared to that for the L513 silent control (P Ͼ 0.05) (Fig. 2F), which may possibly be due to the different genetic background (Dd2 versus 3D7) or the contribution of additional alleles. These data, nevertheless, provide evidence of an association between the PfPFK9 R860G mutation and P. falciparum resistance to PS-3.
Mutations in PfPFK9 lead to changes in central carbon metabolic flux. To investigate the impact of nonsynonymous mutations in PfPFK on parasite metabolism, 13 C-glucose labeling studies were undertaken on 3D7-C3 PS3-1 and 3D7-C3 PS3-3 (both  containing the R860G mutation). Erythrocytes infected with trophozoite-stage parasites were labeled with 13 C-glucose (present in medium at a 1:1 ratio with 12 C-glucose) for 30 min, and incorporation into a wide range of intermediates in central carbon metabolism was quantitated by liquid chromatography-mass spectrometry (LC-MS). The pool sizes of several glycolytic intermediates immediately downstream of PfPFK, including fructose-1,6-bisphosphate and dihydroxyacetone phosphate (DHAP), were significantly reduced in the mutant lines, whereas those of several nucleoside monoand diphosphates were elevated (Fig. 4A). The level of 13 C enrichment in these downstream intermediates was also greatly reduced, suggesting that the metabolic flux through PFK was severely compromised in 3D7-C3 PS3-1 -and 3D7-C3 PS3-3 -infected erythrocytes (Fig. 4B). Reduced flux through the reversible aldolase reaction was further supported by the markedly reduced levels of ϩ3-labeled fructose-1,6-bisphosphate in the mutant lines, which reflects the rate of interconversion of DHAP/glyceraldehyde-3-phosphate (GAP) and fructose-1,6-bisphosphate. These labeling studies indicate that mutations acquired in 3D7-C3 PS3-1 and 3D7-C3 PS3-3 parasites lead to reduced PfPFK activity. Interestingly, 13 C enrichment in GAP, an intermediate in both glycolysis and the pentose phosphate pathway (PPP), was unaltered in 3D7-C3 PS3-1 and 3D7-C3 PS3-3 parasites, while labeling of several intermediates in the oxidative and nonoxidative PPP (sedoheptulose-7-P and ribose-5-P) was increased in these mutant lines. These data strongly suggest that partial loss of PfPFK activity leads to rerouting of carbon flux through the PPP, leading to the production of GAP that can still be catabolized in lower glycolysis. This bypass would allow ATP-producing steps in lower glycolysis to proceed, albeit at a reduced rate, as evidenced by the lower levels of synthesis of phosphoenolpyruvate (PEP) and lactate (Fig. 4B). The reduction of glycolytic flux in 3D7-C3 PS3-1 and 3D7-C3 PS3-3 was consistent with an approximate 25% reduction in extracellular lactate secretion (Fig. 4B). The rerouting of carbon flux through the PPP, at the expense of glycolysis, would be expected to come with a fitness disadvantage as a result of reduced ATP production. This hypothesis was supported by the significant increase in ADP and AMP levels in the mutant lines, leading to reduced ATP/ADP and ATP/AMP ratios (Fig. 4A). Glycolysis is also the source of precursors for a number of anabolic pathways, such as isoprenoid biosynthesis, which uses phosphoenolpyruvate and DHAP generated downstream of PFK. Lastly, the isoprenoid biosynthetic intermediates deoxyribose-1-phosphate (DOXP) and methylerythritol-cyclo-pyrophosphate (MEcPP) were significantly reduced in both 3D7-C3 PS3-1 and 3D7-C3 PS3-3 , indicating reduced flux into this pathway and an additional fitness disadvantage to PFK mutations (Fig. 4C).
3D7-C3 PS3 subclones display slowed in vitro growth compared to that of wild-type 3D7-C3 parasites. To assess in vitro growth dynamics of PS-3-resistant parasites, the in vitro growth profiles of 3D7-C3 PS3-1 and 3D7-C3 PS3-3 , compared to that of wild-type 3D7-C3 parasites, were examined by comparing growth over 72 h commencing at ϳ3 to 6 h postinvasion (see Fig. S4). No significant difference in parasitemia was seen up to 30 h postinvasion for 3D7-C3 PS3-1 and 3D7-C3 PS3-3 compared to that for 3D7-C3. In contrast, at 51 to 54 h and 75 to 78 h postinvasion, a Ͼ2-fold reduction in total parasitemia was observed for both 3D7-C3 PS3 subclones compared to that of the wild type (P Ͻ 0.05 and P Ͻ 0.01, respectively) (Fig. S4). Upon examination of each developmental stage, no significant difference was seen in the percentages of each stage (early rings, mid/late rings, early/mid trophozoites, late trophozoites/early schizonts, and late schizonts) up to 30 h postinvasion for 3D7-C3 versus those for 3D7-C3 PS3-1 and 3D7-C3 PS3-3 (P Ͼ 0.05) (see Fig. S5A and B). However, at 51 to 54 h postinvasion, 2-fold reductions were seen in the percentages of rings for 3D7-C3 PS3-1 and early/mid trophozoites for 3D7-C3 PS3-3 compared to those for 3D7-C3 (P Ͻ 0.05) (Fig. S5C). Additionally, at 72 to 78 h postinvasion, a 2-fold increase in the percentage of late trophozoite/early schizonts and a 5-fold decrease in the percentage of late schizonts was observed for both 3D7-C3 PS3 subclones compared to those for the wild type (P Ͻ 0.01 and P Ͻ 0.05, respectively) (Fig. S5D). Overall, these data suggest that parasite growth is developmentally delayed in 3D7-C3 PS3-1 and 3D7-C3 PS3-3 parasites. . The data presented in panels A and B represent the averages from three biological replicates (Ϯ standard errors of the means [SEMs]) and the excreted lactate data from two biological replicates (ϮSDs). (C) The intracellular pool of the isoprenoid biosynthetic intermediates in wild-type (WT) and 3D7-C3 PS3-1 or 3D7-C3 PS3-3 purified infected erythrocytes. The total pool sizes of deoxyribose-1phosphate (DOXP), methylerythritol-cyclo-pyrophosphate (MEcPP), and isopentyl-pyrophosphate (IPP) are presented as arbitrary ion counts (normalized for data acquisition and degradation across different days), and represent three biological replicates performed on different days (means Ϯ SEMs). One-way analysis of variance (ANOVA) testing was performed to test statistical significance. *, P Ͻ 0.05.
Fisher et al.

Metabolite profiling of 3D7-infected red blood cells treated with PS-3.
To investigate whether PfPFK is a direct target of PS-3 or if mutations in PfPFK provide a metabolic bypass that indirectly confers resistance, erythrocytes infected with purified trophozoite-stage P. falciparum 3D7 were treated with PS-3 (40 M; equivalent to 4ϫ IC 50 for this parasite life cycle stage at 48 h of exposure) for 2 h, and changes in metabolite levels determined using untargeted LC-MS profiling (Fig. 5). More than 90 of the 3,327 detected mass-to-charge (m/z) features were significantly different (P Ͻ 0.05, false-discovery rate [FDR] corrected) between those treated with PS-3 and the untreated control. The METLIN database was queried for possible identities using the M-H criteria with a mass tolerance of 10 ppm (assuming that metabolites existing as other adducts will also be present as the M-H adduct). Several features corresponded to glycolytic intermediates, and these were subsequently verified using authentic standards (Fig. 5, highlighted in yellow). In particular, lactate and pyruvate were significantly increased following PS-3 treatment, with glycerate and the tricarboxylic acid (TCA) cycle-linked intermediates 2-oxoglutarate and 2-hydroxyglutarate also significantly elevated. This metabolic phenotype is distinct from that observed in the PS-3-resistant lines, indicating that PS-3 does not directly target PFK but inhibits other steps in central carbon metabolism that are compensated for by rerouting of glucose into the pentose phosphate pathway and/or reduced flux in lower glycolysis.
Antiplasmodial glycosides do not directly inhibit apicomplexan phosphofructokinases. To further evaluate whether PfPFK9 represents the direct target of antiplasmodial glycosides, we evaluated compound sensitivity against P. falciparum and Plasmodium knowlesi PFK9. While we, and others, have not been able to purify full-length PfPFK9 (5), P. falciparum and P. knowlesi ␣/␤ domains readily express as active individual domains (see Fig. S6). We screened each recombinant protein for enzymatic inhibition at a single compound concentration (50 M) (Fig. 6A), revealing possible modest inhibition of purified PfPFK␤ by compounds PS-3 and PS-12. However, dose-responsive inhibition of PfPFK␤ confirmed little to no enzymatic inhibition by these compounds at concentrations similar to the compound's cellular inhibition values (biochemical IC 50 s as follows: PS-3, Ͼ22 Ϯ 4.33 M; PS-12, Ͼ21 Ϯ 1.49 M) (Fig. 6B). The lack of direct inhibition of any domain of these PfPFK enzymes suggest PfPFK9 may not be the direct target of PS-3.  Fig. S7). This lack of cross-resistance suggests that PS-3 may not target PfFNT.
PS-3-resistant clones are hypersensitive to fosmidomycin. Our metabolomic studies indicated that glycolytic flux is reduced in 3D7-C3 PS3 subclones, which leads to decreased flux of glycolytic intermediates into the apicoplast isoprenoid DOXP biosynthetic pathway (Fig. 4C). Previous studies on the PfHAD1 mutants have shown that changes in glycolytic flux directly impact the resistance of asexual intraerythrocyticstage P. falciparum to fosmidomycin (9), a potent competitive inhibitor of the apicoplast MEP pathway enzyme deoxyxyluose phosphate reductoisomerase (PfDXR) (19). We therefore assessed whether the observed decrease in glycolytic flux in the PS-3resistant lines was associated with increased sensitivity to fosmidomycin in 72-h growth inhibition assays. A significant increase in activity for fosmidomycin was seen for all three 3D7-C3 PS3 clones compared to that for the 3D7-C3 wild-type parasites (Fig. 7A)   ). As previously observed (Fig. 2C), PS-3 showed a significant decrease in activity for all three 3D7-C3 PS3 clones compared to the that for 3D7-C3 (Fig. 7B) (ϳ7-to 15-fold lower IC 50 s; P Ͻ 0.001). The PS-3-resistant parental line (3D7-C3 PS3 ) also displayed hypersensitivity to the P. falciparum apicoplast inhibitor clindamycin (see Fig. S8) (ϳ3-fold higher IC 50 values for 3D7-C3 compared to that for 3D7-C3 PS3 ; n ϭ 2). Collectively, these data support the hypothesis that resistance to PS-3 in the 3D7-C3 PS3 clones is associated with metabolic adaptation that leads to decreased availability of glycolytic intermediates for anabolic pathways, such as isoprenoid biosynthesis.
Increased glycolytic flux may also result in resistance to PS-3. To further investigate the mode of action of PS-3 and potential resistance mechanisms, we investigated the sensitivity of the P. falciparum had1 mutant, AM1, to PS-3. The Pfhad1 gene encodes a haloacid dehalogenase that dephosphorylates a range of glycolytic intermediates in vitro and appears to be involved in negatively regulating glycolytic flux. Loss-of-function mutations in Pfhad1 are associated with increased flux of glycolytic intermediates into isoprenoid synthesis and resistance to fosmidomycin (9). The AM1 parasite line exhibited a 5-fold increase in resistance to PS-3 compared to that of the parental 3D7 line (P Ͻ 0.001) (Fig. 8A). This level of resistance is comparable to that exhibited by the AM1 line to fosmidomycin (i.e., 4-fold increase in IC 50 ; P Ͻ 0.05) (Fig. 8B). This finding suggests that global changes in central carbon metabolism that can be associated with either an increase in glycolytic flux (as occurs in AM1) or a decrease in glycolytic flux (as occurs in the 3D7-C3 PS3 subclones) increase the resistance of asexual stages to PS-3. However, unlike the control fosmidomycin, the antiplasmodial activity of PS-3 was not rescued by supplementation with isoprenoid isopentenyl pyrophosphate (IPP) (see Fig. S9). Furthermore, we find that PS-3 does not directly inhibit PfHAD1 or PfHAD2 enzyme activity in vitro (Fig. 9). Together, these data strongly show that PS-3 has a distinct mechanism of action from other direct MEP pathway inhibitors, such as fosmidomycin. falciparum Fos r 3D7-AM1 and wild-type 3D7-MR4 against compound PS-3 (A) and fosmidomycin (B) was assessed using 72-h [ 3 H]hypoxanthine uptake growth inhibition assays. In each case, three independent assays, each in triplicate wells, were carried out, and 50% inhibitory concentrations (IC 50 ) were determined using nonlinear regression analysis in GraphPad prism.

DISCUSSION
Malaria remains a major global concern with efforts to control this disease being hampered by parasite drug resistance and the lack of a broadly effective vaccine (1). Although there are several new chemotherapeutics in various stages of preclinical or clinical development, most of these have previously exploited targets or are reformulations or different combinations of existing antimalarial drugs (20). Therefore, strategies to identify new antimalarial drug targets and antiplasmodial chemotypes that have novel modes of action are crucial. Fortunately, recent reductions in the cost of wholegenome sequencing and the development of comprehensive global metabolomic profiling approaches have begun to yield some exciting progress in the antimalarial target identification arena. When combined with the selection of drug-resistant P. falciparum lines, this approach has resulted in the identification of new antimalarial targets and resistance mechanisms (11,21,22).
In this study, in vitro resistance selection was utilized to generate P. falciparum parasites resistant to an antiplasmodial glycoside (PS-3) (10). Importantly, PS-3-resistant parasites displayed no in vitro resistance to several clinically used antimalarial drugs or to two compounds that represent chemical classes under clinical development (11,12), suggesting a novel mode of action/resistance. Historically, the PS moiety is known to inhibit carbonic anhydrase (CA) enzyme activity in many organisms (23), including P. falciparum (24,25). In this study, we demonstrate that PS-3 resistance in P. falciparum is not linked to the PS moiety, as no change in the resistance profile was observed for PS-3=, the non-PS structural analogue of PS-3 (Table 2). Further screening of PS glycoside analogues against PS-3-resistant parasites revealed that the glucose component of PS-3 may be contributing to the resistance phenotype. These data support genome sequencing data of 3D7-C3 PS3 clones and subsequent validation using reverse genetics, which indicate that the point mutation (R860G) in PfPFK9, the gene that encodes an isoform of the glycolytic enzyme phosphofructokinase, contributes to PS-3 resistance. However, the resistance index of the R860G-edited line was lower than that observed for the 3D7-C3 PS3-1 and 3D7-C3 PS3-3 clones (Ri, ϳ2.1, 10.8, and 15, respectively). This may be attributed to mutations in the P. falciparum multidrug-resistant protein 2 (PfMDR2; Pf3D7_1447900) and/or the P. falciparum sodium/hydrogen exchanger (PfNHE; Pf3D7_1303500), observed in the 3D7-C3 PS3-1 and 3D7-C3 PS3-3 clones only (see Table S1 in the supplemental material). The role of PfMDR2 in antimalarial drug resistance is unclear, with some weak evidence linking resistance to pyrimethamine (26) and sulfadoxine (27) treatments. Likewise, there are contradictory reports on the association of sequence polymorphisms in the PfNHE gene and quinine resistance (28)(29)(30)(31). In this study, the 3D7-C3 PS3 parasites were not resistant to pyrimethamine and quinine (Table 1); however, the PfMDR2 and PfNHE mutations observed in the 3D7-C3 PS3-1 and 3D7-C3 PS3-3 clones were not reported previously; therefore, further studies are required to confirm if these mutations are associated with resistance to PS-3. While mutations in PfMDR2 and PfNHE genes may be contributing to PS-3 resistance, this is more likely to be via nonspecific multidrug resistance mechanisms (32,33). On the other hand, the PFK enzyme has an essential housekeeping role in P. falciparum central carbon metabolism; therefore, mutations in this gene are more likely to be target associated. Surprisingly, the L513Q-edited line was not resistant to PS-3, and the 3D7-C3 PS3-2 clone containing this mutation did not share any other common mutations with the other two clones (Table S1). In this case, it is possible that multiple background mutations (Table S1) are contributing to the resistance profile of the 3D7-C3 PS3-2 clone and masking the effect that the L513Q mutation may be exhibiting. Alternatively, our modeling suggests that the L513Q mutation may have less of an effect on PfPFK function due to its more distal proximity to the substrate-binding pocket than the R860G mutation (Fig. 3). It should also be noted that the L513Q clone reported in this study is a subclone of the original L513Q clone generated. This original clone did display resistance to PS-3 (ϳ6-fold increase in IC 50 [n ϭ 1] for two independent PS-3 stocks; data not shown). However, this phenotype was lost following cryopreservation, and a reversion to the wild-type genotype was observed. It is therefore conceivable that the L513Q subclone may have gained mutations in another region of the genome that may have restored sensitivity to PS-3. Nonetheless, overall, these data strongly suggest that the PfPFK R860G mutation contributes to PS-3 resistance, suggesting that alterations in glycolytic flux help to bypass the mode of action of PS-3.
To further understand the impact that the R860G PfPFK mutation has on P. falciparum metabolism, 13 C-glucose fluxes were measured in both the PS-3-resistant clones (3D7-C3 PS3-1 and 3D7-C3 PS3-3 ) and 3D7-C3 wild-type parasites. These studies revealed a profound rewiring of glucose fluxes in the two resistant clones. In particular, a reduction in labeling of intermediates immediately downstream of PFK was associated with increased flux of 13 C-glucose into the PPP (Fig. 4). The redirection of glucose-6phosphate into the PPP likely accounts for the reduction of ATP/ADP and ATP/AMP ratios and reduced in vitro growth of PS-3-resistant parasites (Fig. S4). On the other hand, increased flux into the PPP would allow regeneration of NADPH and contribute to a more robust redox state. Surprisingly, our data suggest that PfPFK may not be the direct target of PS-3, as PS-3 was unable to inhibit recombinant PfPFK and P. knowlesi PFK (PkPFK) enzymes (Fig. 6). Furthermore, in silico homology modeling predicts that PS-3 is not likely to affect PFK substrate binding or catalysis. Nonetheless, it should be noted that PS-3 was only tested against the individual subunits of PfPFK, as no full-length enzyme is available; thus, we cannot completely rule out PfPFK9 as a target of PS-3. However, PS-3 could in principal inhibit another enzyme in the glycolytic pathway or in the pentose phosphate pathway. Initially, we hypothesized that PS-3 may inhibit the parasite's lactate transporter (P. falciparum formate nitrite transporter, or PfFNT, a validated drug target [16,17]), resulting in a toxic buildup of metabolites. This would be consistent with the increases seen in the lower glycolytic intermediates (lactate, pyruvate, and glycerate) in PS-3-treated trophozoites (Fig. 10). PS-3-resistant parasites may overcome this toxicity by redirecting glycolytic flux away from lower Phosphofructokinase in Glycoside Resistance ® glycolysis, resulting in less lactate being excreted and hence reducing the potency of PS-3. However, the L513Q and R860G mutant parasites were not cross resistant to the PfFNT inhibitors MMV000972 and MMV007839 (18) (Fig. S7). Moreover, the elevation of pyruvate and lactate following PS-3 treatment is consistent with PfFNT inhibition, but PS-3 treatment does not lead to the additional perturbations reported for PfFNT inhibition (e.g., pyrimidine biosynthesis and hemoglobin catabolism [18]). However, it is possible that differences in lactate accumulation may exist between Pf3D7 and PfDd2, warranting PfFNT inhibition studies against the 3D7 PS-3-selected clones. To the best of our knowledge, there are no reports comparing 3D7 and Dd2 lactate metabolism, but we do know that 3D7 lactate metabolic profiles are similar to the profiles of other multidrug-resistant strains 7G8 and K1 (34) and a Pf3D7 chloroquine-resistant transporter (PfCRT) mutant line (35). Therefore, while we cannot completely discount the possibility that PS-3 may partially block PfFNT, with the evidence presented, we think another target is leading to the observed disruptions and eventual cell death. For example, PS-3 may inhibit an enzyme in the PPP, accounting for the accumulation of intermediates in glycolysis (glucose-6-P, pyruvate, lactate) and the TCA cycle (2oxoglutarate and 2-hydroxyglutarate). In this case, PS-3-resistant parasites may have adapted by diverting the glycolytic intermediates (fructose-6-phosphate [F6P] and GAP) into the nonoxidative arm of the pentose phosphate pathway to overcome the effects of PS-3 (Fig. 10). In P. falciparum, the nonoxidative reactions of the PPP are mainly designed to produce ribose-5-phosphate (R5P) for nucleic acid synthesis. The nonoxidative arm of the PPP can also utilize the glycolytic intermediates (F6P and GAP) to produce R5P and vice versa (36). Two key enzymes involved in these processes are glucose-6-phosphate dehydrogenase 6-phosphogluconolactonase (PfGluPho) (37) and transketolase (38), both of which have been characterized, are essential for asexual growth (6,37), and differ structurally from their human homologues (37,38), making them potential drug targets. Finally, it is possible that PS-3 may target other processes that lead to increased oxidative stress or loss of redox balance, such as mitochondrial metabolism, which is compensated for by increased flux through the PPP (with concomitant regeneration of NADPH).
In addition to reduced glycolytic flux, we also observed a reduction in the MEP pathway metabolites (DOXP, MEcPP, and IPP) in PS-3-resistant parasites (Fig. 10). The MEP pathway is essential for the synthesis of isoprenoids in P. falciparum (39) and is dependent on the continuous import of glycolytic intermediates from the cytoplasm (9). The lead antimalarial drug fosmidomycin targets the MEP pathway enzyme, deoxyxylulose phosphate reductoisomerase (DXR), and fosmidomycin resistance has been linked to mutations in P. falciparum haloacid dehalogenases (PfHAD1/2) (7,9). While we found that fosmidomycin-resistant parasites (Fos r AM1 PfHAD1 mutant) were also resistant to PS-3, the activity of PS-3 was not able to be rescued by IPP (isoprenoid precursor), and PS-3 did not inhibit Plasmodium HAD1/HAD2 recombinant enzymes. This suggests that PS-3 has a different mechanism of action/resistance to fosmidomycin. In contrast, PS-3-resistant parasites were found to be hypersensitive to fosmidomycin. Given that fosmidomycin targets DXR, it is likely that the reduction in DXR's substrate (DOXP) in PS-3-resistant parasites, coupled with a decrease in MEcPP and IPP, results in an increase in fosmidomycin activity. Increased sensitivity to fosmidomycin has also been reported in P. falciparum parasites lacking phosphoglycolate phosphatase (PGP) (40). PGP, a third member of the P. falciparum HAD family, has been shown to be involved in regulating glycolysis and PPP flux in asexual P. falciparum (40). In Δpgp parasites, the loss of PGP leads to the inhibition of the PPP enzyme 6-phosphogluconate dehydrogenase (6-PGD), resulting in reduced glycolytic flux and causing reduced isoprenoid biosynthesis and increased sensitivity to fosmidomycin (40). Interestingly, fosmidomycin-resistant parasites harboring mutations in Pf-HAD2 display the opposite metabolomic profile to that of PS-3-resistant parasites. As mentioned above, PS-3-resistant parasites exhibit decreased levels of the metabolites FBP, DOXP, and MEcPP, whereas a reported increase in these metabolites is seen in PfHAD2 mutants (7). In PfHAD2 mutants, the increase in FBP is thought to lead to the increase of glycolytic intermediates into the MEP pathway, which results in an increase of MEP pathway metabolites, thus reducing the sensitivity to fosmidomycin (7). Furthermore, mutations in PfPFK9 have been shown to restore fosmidomycin sensitivity in these PfHAD2 mutant parasites (7); however, these mutations are not the same as those seen in PS-3-resistant parasites. Overall, these findings provide further evidence that PfPFK9 plays an important regulatory role in central carbon metabolism to modify drug sensitivity.
In this study, we identified a novel mechanism of P. falciparum drug resistance and have further highlighted the importance of the role of PfPFK9 in regulating central carbon metabolism to offset the effects of antiplasmodial compounds. While the specific molecular target of PS-3 remains unclear, it is promising that no cross-resistance exists with several of our clinically used antimalarials, indicating that this chemical class is likely to have a different mode of action from that of our current drugs. Furthermore, our PS-3-resistant parasites will provide a useful tool for studying central carbon metabolism in P. falciparum and aid in identifying other inhibitors of these pathways for development as antimalarial drug leads.

MATERIALS AND METHODS
Compounds. PS glycosides and matched control compounds were synthesized as previously reported by us (10). Chloroquine, artesunate, pyrimethamine, and cytochalasin B were purchased from Sigma-Aldrich, USA. DSM161 (12) was supplied by Margaret Philips from UT Southwestern, Dallas, TX, USA. Stock solutions were prepared in 100% dimethyl sulfoxide (DMSO; Sigma-Aldrich, USA), stored at Ϫ20°C and diluted as required.
Cloning of P. falciparum parasites by limiting dilution. P. falciparum-infected erythrocytes were cultured under standard culture conditions and synchronized to ring stage using sorbitol treatment (42). Synchronous ring-stage parasites were cultured for two cycles (96 h) on a plate shaker (Gyro mini; Labnet, USA) under standard culture conditions until multiple infections were less than 1%, as determined by microscopic examination of Quickdip-stained (POCD, Australia) thin blood films. Following cell counts with a hemocytometer, cultures were diluted to 0.5 and 0.1 parasites per 200 l (2% hematocrit), dispensed into sterile 96-well plates (Corning, USA), and incubated under standard culture conditions with medium replaced weekly. On day 16, stained thin blood smears were prepared and stained with Quickdip (POCD) and then examined by microscopy. Parasite-positive cultures were transferred into 50-mm petri dishes (Corning) and then 100-mm petri dishes (Corning) for expansion and cryopreservation. The parasite-negative wells were checked 1 week later by microscopic analysis of Quickdip-stained blood smears, and no further parasite-positive wells were identified.
In vitro P. falciparum growth inhibition assays. In vitro inhibition of P. falciparum growth was assessed using a 72-h isotopic microtest, essentially as previously described (43). Briefly, highly synchronous ring-stage P. falciparum-infected erythrocytes obtained by sorbitol treatment (42) were seeded at 0.5% parasitemia and 2.5% final hematocrit into 96-well tissue culture plates (3596 Costar; Corning, USA) containing serial dilutions of control or test compounds. Compound vehicle only (0.5% final DMSO) and the antimalarial drug chloroquine served as negative and positive controls, respectively, in each assay. After incubating for 48 h under standard P. falciparum culture conditions in RPMI 1640 medium (Gibco, USA) containing 10% heat-inactivated pooled human sera and 5 g/ml gentamicin (Sigma, USA), 0.5 Ci [ 3 H]-hypoxanthine (PerkinElmer, USA) was added to each well followed by culturing for a further 24 h. Cells were harvested onto 1450 MicroBeta filter mats (Wallac, USA), and 3 H incorporation was determined using a 1450 MicroBeta liquid scintillation counter (PerkinElmer). Percentage inhibition of growth for compound treated versus that for matched vehicle only (0.5% DMSO) controls was determined, and IC 50 values were calculated using nonlinear regression analysis in GraphPad Prism. Each compound was assayed in triplicate wells in at least three independent experiments. Statistical difference between IC 50 s was determined using an unpaired t test with GraphPad Prism data analysis software.
In vitro P. falciparum resistance selection. P. falciparum 3D7 clone C3 (3D7-C3)-infected erythrocytes were cultured with (and, in parallel, without) PS-3 at 1ϫ IC 50 (0.9 M). Parasite growth was monitored every 1 to 3 days via microscopic examination of Quickdip-stained (POCD, Australia) thin blood films, with medium changed as needed. When PS-3-treated parasites were observed to be replicating at a similar rate to that of the untreated controls, compound pressure was gradually increased in a stepwise manner over several weeks until the parasites were surviving in 10ϫ IC 50 concentrations. At this point, the selected and wild-type clones were assessed in growth inhibition assays to determine sensitivity for PS-3 and other compounds. Once a resistance phenotype was confirmed, PS-3-selected and wild-type parasites were subcloned and phenotype assessed again to confirm resistance.
In vitro growth rate analysis of P. falciparum. Highly synchronous ring-stage P. falciparum cultures starting at 0.25% rings and 2.5% hematocrit were cultured under standard conditions for 72 h. Quickdip-stained (POCD, Australia) thin blood films were prepared every 24 h, and ϳ3,000 erythrocytes Phosphofructokinase in Glycoside Resistance ® were counted by two independent microscopists to determine the mean number of parasites infecting 100 erythrocytes at each time point. Three independent assays were assessed per clone.
Genome sequencing. Genomic DNA was isolated from P. falciparum-infected erythrocytes using a DNeasy blood and tissue kit (Qiagen, USA). The Nextera XT kit (Illumina) was used to prepare DNA libraries from samples for whole-genome sequencing using the dual index protocol. The libraries were run on the Illumina HiSeq 2500 in rapid run mode with 100-bp paired-end reads. The reads were aligned to the P. falciparum 3D7 reference genome (PlasmoDB v. 13.0) as described previously (44). Single nucleotide polymorphisms (SNPs) and indels were called with the Genome Analysis Toolkit's (GATK) HaplotypeCaller (45,46). Variants were filtered by quality scores and sequencing bias statistics based on GATK's default filtering parameters. SNPs were filtered out if they met any of the following criteria: quality depth (QD), Ͻ2.0; mapping quality (MQ), Ͻ50.0, Phred-scaled P value using Fisher's exact test to detect strand bias (FS), Ͼ60.0; symmetric odds ratio (SOR), Ͼ4.0; Z-score from Wilcoxon rank sum test of alternative versus reference read mapping qualities (MQRankSum), less than Ϫ12.5; ReadPosRankSum (RPRS) parameter, less than Ϫ8.0. Indels were filtered out if they met any of the following criteria: QD, Ͻ2.0; RPRS, less than Ϫ20.0; FS, Ͼ200.0. Variants were annotated using snpeff (version 4.2) (47). Custom scripts were used to compare the variants between the parent sequence and the resistant clones.
CRISPR-Cas9 genome editing. CRISPR/Cas9 editing of mutations in PfPFK9 was performed using a pDC2-based Cas9 guide RNA (gRNA) plasmid, pDC2-cam-coCas9-U6.2-hDHFR. Two sgRNAs were designed per target site using Benchling (San Francisco, CA). For targeting of L513, gRNA1 (CAATTTATGT CACATTATCT) and gRNA2 (TCACATTATCTAGGTTATGA) were employed, and for targeting of R860, gRNA3 (CATAACACATTCATAGCACC) and gRNA4 (GGTGCTATGAATGTGTTATG) were used. Donor templates with 656-bp or 750-bp homology to the L513 and R860 target sites, respectively, were synthesized (Thermo Fisher) and cloned into the AatII-EcoRI sites of the Cas9 vector using Gibson assembly. Plasmids were transfected by electroporation (0.31 kV, 950 F) into Dd2 parasites and selected with 2.5 nM WR99210 for 8 days before drug pressure was removed. Editing of the recovered parasites was examined by Sanger sequencing of the bulk culture, and clonal lines were derived by limiting dilution.
Stable-isotope incorporation was performed using methods previously described with minor modifications (48). Briefly, purified infected erythrocytes were resuspended in fresh RPMI 1640 medium at a cellular density of 1 ϫ 10 8 cells/ml and allowed to recover for 10 min. Time courses were initiated by adding an equal volume of RPMI 1640 containing 11 mM 13 C-U-glucose (Sigma), leading to a 1:1 mix of fully unlabeled/fully labeled glucose. This 1:1 mixing was performed to avoid perturbation of metabolism during label addition and allows detection of more complex labeling patterns. At predetermined time points, 1 ϫ 10 8 cells were aliquoted, centrifuged (15 s at 14,000 ϫ g), washed with 1 ml ice-cold phosphate-buffered saline (PBS), and centrifuged (15 s at 14,000 ϫ g), and metabolites were extracted with 200 l of 80% acetonitrile (containing 1 M 13 C-U-aspartate). Samples were rapidly vortexed and centrifuged (5 min 14,000 ϫ g), and the supernatant was collected.
The metabolites were separated on a SeQuant ZIC-pHILIC column (5 M, 150 mm by 4.6 mm; Millipore) with a 1260 series high-pressure liquid chromatography (HPLC) system (Agilent) using a method previously described with modifications (49). Briefly, a flow rate of 0.3 ml/min was used with 20 mM ammonium carbonate in water (A) and 100% acetonitrile (B) as the mobile phase. A binary gradient was set up as follows: 0.5 min, 80% B; 15.5 min, 50% B; 17.5 min, 20% B; 18.5 min, 5% B; 21 min, 5% B; 23 min, 80% B; held at 80% B until 29.5 min. Detection of metabolites was performed on an Agilent Q-TOF mass spectrometer 6545 operating in negative electrospray ionization (ESI) mode. The scan range was 85 to 1,200 m/z between 2 and 27 min at 0.8 spectra/second. LC-MS .d files were converted to .mzXML files using MS convert and analyzed using MAVEN (50). Following alignment, metabolites were assigned using exact mass (Ͻ10 ppm) and retention time (compared to a standards library of 150 compounds run the same day). Isotopologues for each metabolite of interest were extracted and integrated, and percent 13 C incorporation was converted into concentration of metabolite labeled using the absolute metabolite concentrations determined previously (48). When metabolite values were not previously determined (ribose-5-P, sedoheptulose-7-P, octulose-8-P, and malate), metabolite concentrations were arbitrarily set to 100 M.
The drug-induced changes to the parasite metabolite profile were determined using the experimental approach described in reference 49. Drug was added at 40 M to cell suspensions (each containing 1 ϫ 10 8 cells at 0.4% hematocrit), which were incubated at 37°C under controlled atmospheric conditions (5% CO 2 and 1% O 2 in N 2 ). After 2 h of incubation, cell suspensions were processed and metabolites were extracted for LC-MS analysis as described above.
Isopentenyl pyrophosphate rescue. In vitro isopentenyl pyrophosphate (IPP) pathway rescue against P. falciparum 3D7 parasites was carried out as previously described (39). Briefly, PS-3 was tested in P. falciparum in a 72-h [ 3 H]hypoxanthine growth inhibition assay as previously described (43), with the following modifications. Two identical 72-h assays were performed simultaneously, one supplemented with 200 M IPP and one without IPP. The antibiotic antimalarial compound clindamycin was used as a positive control. Three or more independent assays were performed, each in triplicate wells. Each compound was assayed in triplicate wells in at least three independent experiments. Statistical difference between IC 50 s was determined using a two-tailed t test with GraphPad Prism data analysis software.