Genetic mapping of targets mediating differential chemical phenotypes in Plasmodium falciparum.

Studies of gene function and molecular mechanisms in Plasmodium falciparum are hampered by difficulties in characterizing and measuring phenotypic differences between individual parasites. We screened seven parasite lines for differences in responses to 1,279 bioactive chemicals. Hundreds of compounds were active in inhibiting parasite growth; 607 differential chemical phenotypes, defined as pairwise IC(50) differences of fivefold or more between parasite lines, were cataloged. We mapped major determinants for three differential chemical phenotypes between the parents of a genetic cross, and we identified target genes by fine mapping and testing the responses of parasites in which candidate genes were genetically replaced with mutant alleles. Differential sensitivity to dihydroergotamine methanesulfonate (1), a serotonin receptor antagonist, was mapped to a gene encoding the homolog of human P-glycoprotein (PfPgh-1). This study identifies new leads for antimalarial drugs and demonstrates the utility of a high-throughput chemical genomic strategy for studying malaria traits.

Malaria is a serious public health burden that causes an estimated 1–2 millions deaths and 300–500 million infections each year1. There is no effective vaccine available, and parasites resistant to almost all antimalarial drugs currently in use have been reported, including reduced sensitivity to derivatives of the traditional Chinese medicine qinghaosu (artemisinin, 2)2. Development of new drugs and a better understanding of the targets of antimalarial drugs and drug resistance are urgently needed.

Phenotypic characterization of human malaria parasites is limited because pathogenic stages live within red blood cells (RBCs) and laboratory models for in vivo investigations are often unsatisfactory. Changes in response to antimalarial drugs, differences in growth rate, or variations in virulence are among the few phenotypes typically accessible3. When phenotypes are available, genetic mapping is a powerful tool to assign them to particular determinants, and various high-throughput genotyping methods, including microarrays for detecting single nucleotide polymorphism (SNPs) and microsatellites (MS), have been developed for studies of Plasmodium falciparum traits4. However, there are ~5,400 predicted genes in the parasite genome, and the function of the majority of these genes remains unknown5. Characterizing phenotypic differences in malaria parasites and identifying the genes affecting the differences may provide important information for investigating gene function.

A challenge in understanding drug action and mechanisms of drug resistance is to identify the relevant molecular target. One useful strategy is to synthesize an active compound and use it to affinity purify the protein target(s) to which the compound binds6. This approach, however, generally requires compounds that have high affinity for their targets. Another strategy employs genetic mapping to link chromosomal loci that affect parasite responses to compounds, allowing molecular targets to be identified after fine mapping and functional characterizations of candidate genes. In addition to discovering potential new antimalarial compounds, these strategies can detect and define differential chemical phenotypes (DCPs) that show distinct signature responses to compounds among a variety of parasite isolates.

Here we demonstrate a strategy for identifying targets of chemical compounds in malaria parasites by integrating quantitative high-throughput screening (qHTS) with genetic mapping (Fig. 1). We tested seven P. falciparum lines, including parents of three genetic crosses7–9, for their responses to 1,279 bioactive compounds and identified candidate genes for three DCPs using progeny from a genetic cross and genetic transfection methods of allelic replacement. These results show that differential responses of small molecules between parasite lines can be reliable phenotypes for exploring molecular mechanisms of pharmacologic interest. This study also provides an effective approach for investigating drug action and resistance mechanisms in diseases other than malaria.

Figure 1

A chemical genomic strategy for studying gene function in malaria parasites. Quantitative high-throughput screening (qHTS) of parasites against compounds in titration-response fashion identifies a large number of differential chemical phenotypes (DCPs). Target genes associated with these DCPs can be identified using quantitative trait loci analysis after genotyping progeny from genetic crosses or field isolates. Gene functions can be deduced from classes of compounds that target specific biologic pathways. The green circles represent differential parasite responses to chemicals. C1 and C2 represent negative and positive controls; and P1–P3 represent responses from three parasites. Gametocytes are the sexual stage of malaria parasite that can be cultured in vitro, and a genetic cross is started by feeding a mixture of gametocytes from two different parasites to mosquitoes.

RESULTS

Quantitative high-throughput screening of bioactive compounds using a Plasmodium proliferation assay

We tested P. falciparum proliferation within infected erythrocytes against the LOPAC1280 collection of known bioactives (Sigma-Aldrich) by a qHTS 10 of a SYBR DNA binding assay11. Parasite proliferation was measured after 72 hr of incubation (corresponding to 1.5 generations of intra-erythrocytic parasite growth), with each compound tested at seven or eight five-fold dilutions beginning at 29 μM. Two independent screens of each parasite line performed well, showing 0.7 or higher average Z' factor and eight-fold or higher signal-to-background ratio (Supplementary Table 1). The potencies of known antimalarial agents had similar values determined by the assay in 96-well plate format. Titrations of two control inhibitors, 2 and mefloquine (3), were present on every plate and showed expected IC50 values (Supplementary Table 1). The antimalarial agents chloroquine (4) and quinine (5) were present in the collection, and IC50 values determined from the qHTS for 5 were similar, but the measurements for 4 were 15- to 20-fold higher than those from 96-well plate tests (Supplementary Table 1) because of lower solubility of 4 in dimethyl sulfoxide (DMSO) (data not shown). Although the determined potencies of 4 were lower, the relative potency between lines sensitive to 4, HB3, 3D7, and D10 were clearly distinguished from resistant lines. The consensus IC50 and activity values for each of the 1,279 compounds in each of the seven P. falciparum lines are listed in Supplementary Table 2 and screening data are deposited in PubChem (AID1828). Comparison of the replicate runs for each parasite line indicated excellent agreement of curve class assignment and IC50 determination. About 80% of actives (compounds associated with Class 1.1, 1.2, and 2.1 curves10; see Methods for curve class definitions) identified in one replicate were active in the second replicate for all lines except Dd2, where 55% were active in both replicates. Of the actives that did not repeat, almost all showed inconclusive activity (Class 2.2 and 3) in the other replicate with few or none scoring as inactive (Supplementary Table 3). The potencies of compounds scored as active or inconclusive in both replicates correlated well (Supplementary Fig. 1), indicating good repeatability in determining IC50 values between replicates.

Discovery of potential antimalarial compounds

Screens of the seven parasite lines revealed a large number of consensus actives (active in both replicates or active in one replicate and inconclusive in the other), all of which inhibited parasite growth. Among the 1,279 compounds tested, about 20% to 30% were active in most lines except W2, where 40% were active, and D10 and Dd2, where 19% and 15% were active, respectively (Supplementary Table 3). Of the hundreds of inhibitors identified for each line, about 50% or more showed IC50 values between 1 and 10 μM and 6% to 14% had IC50 values less than 1 μM (Supplementary Table 4). There were 155 compounds that inhibited growth in all seven lines tested (Supplementary Table 5). The potency distribution of these pan inhibitors indicated differences in sensitivity between the lines; W2 was most sensitive, with 32% of the compounds having IC50 values of 1 μM or less, while Dd2 was least sensitive with 7% below 1 μM. We identified 25 potent compounds that inhibited proliferation in all parasite lines at lower than 2 μM IC50 (Supplementary Table 6). Some of these compounds are well known antimalarial drugs (5 and quinacrine [6]), while others such as hexahydro-sila-difenidol hydrochloride (7), dequalinium dichloride (8), taxol (9), and BW 284c51 (10) are not compounds used for malaria prophylaxis or treatment.

Identification of a large number of DCPs

One of our study objectives was to identify and profile differences in parasite responses to different chemicals. These differences can be considered phenotypes that derive from underlying variations in each parasite genotype. Hierarchical clustering of activity category (Fig. 2a) and IC50 values (Fig. 2b) of actives showed clear differences in the parasite responses to many compounds. We therefore compared compounds between each pair of parasite lines that showed a five-fold or greater difference in IC50 (see Methods for specific criteria). Applying these criteria, we identified 149 compounds (Supplementary Table 7) that produced 607 potential DCPs between the seven lines (Table 1). Interestingly, the three chloroquine-sensitive (CQS) parasite pairs had the smallest differences in response to the chemicals: seven DCPs for HB3 and 3D7, two for HB3 and D10, and seven for 3D7 and D10. There were 58 DCPs between the parents of the three genetic crosses, with seven between 3D7 and HB3, 23 between GB4 and 7G8, and 28 between Dd2 and HB3 (Table 1). These DCPs can be used to identify responsible loci or genes via linkage mapping of recombinant progeny from the crosses.

Figure 2

Hierarchical clustering of compound activities in seven Plasmodium falciparum lines. (a) Clustering of compounds based on activity category from parasite responses to the compounds. Each compound was scored as consensus active (Class 1.1, 1.2, or 2.1 in one or both replicates; red), consensus inconclusive (Class 2.2 or 3 in both replicates; pink) or consensus not active (Class 4 in one or both replicates; white) for each parasite line. Structures of four known dihydrofolate reductase inhibitors and their positions in the clustering are shown. (b) Hierarchical clustering of IC50 values of differentially active compounds in seven P. falciparum lines. IC50 values of 149 compounds having five-fold or greater potency differences between two or more strains were clustered. IC50 values ranged from 12 nM (dark red) to > 29 μM (inactive; white). Grey indicates excluded IC50 values

Table 1

Number of potential differential chemical phenotypes from compounds with five-fold differences in IC50 values between parasite pairs

Parasite	3D7	7G8	D10	Dd2	GB4	HB3	W2
3D7	0
7G8	23	0
D10	7	34	0
Dd2	33	21	31	0
GB4	27	23	39	33	0
HB3	7	18	2	28	26	0
W2	39	21	74	53	16	52	0

Mapping molecular targets underlying selected DCPs

To confirm that genes underlying DCPs could be identified using genetic mapping, we examined (1), trimethoprim (11), and triamterene [12]) because they exhibited relatively large differences in IC50 between the parents of a GB4×7G8 cross. 1 (Fig. 3a) is a competitive serotonin receptor antagonist and has been used to treat migraine headache12,13. From the qHTS, 1 was ten times more potent in 7G8 (0.2 μM) compared with GB4 (2 μM). Follow-up tests in 96-well plates confirmed the IC50 difference between 7G8 (0.37 ± 0.01 μM) and GB4 (2.8 ± 0.37 μM; Fig. 3b). We then tested the 1 responses of 32 progeny from the cross and used the resulting IC50 values and genotypic data from 285 MS markers9 to map genetic loci controlling the difference in parasite responses (Fig. 3b). Quantitative trait loci (QTL) analysis identified a major locus on chromosome 5 with an LOD score of 16.4 and another locus on chromosome 12 having a LOD score of 3.0 (Fig. 3c). A perfect genotype-phenotype association was found within a ~150-kb segment of chromosome 5 bounded by crossovers at marker B7M114 in progeny JB8 and marker B5M86 in progeny XE7, LA10, and WE2 (Fig. 3b). This 150-kb segment contained pfmdr1, the gene encoding a homolog of the human P-glycoprotein (PfPgh-1) involved in drug resistance, and 33 other predicted genes (Supplementary Table 8). The smaller peak on chromosome 12 suggests that a second gene might modulate the response to 1 (Fig. 3c).

Figure 3

Mapping genetic loci contributing to IC50 differences between 7G8 and GB4 in response to dihydroergotamine methanesulfonate (1). (a) Chemical structure of 1. (b) IC50 values (mean and standard deviation) of 1 and allelic designations of three microsatellite markers on chromosome 5 and one at the gene encoding the chloroquine resistant transporter (pfcrt) on chromosome 7 are shown for the parents and progeny of a GB4×7G8 cross. A perfect match of sensitivity to 1 and microsatellite polymorphism in pfmdr1 is seen. (c) Peaks of quantitative trait loci analysis linked to differential 1 responses are shown. Predicted genes within the chromosome 5 locus can be found in Supplementary Table 8.

11 (Fig. 4a) is an antifolate that targets DHFR and is used mainly for prophylaxis and treatment of urinary tract infections14. In follow up assays, the parasite clone 7G8 had a much higher IC50 (37 ± 1.2 μM) to 11 than did GB4 (2.5 ± 0.2 μM). Analysis of responses of 32 GB4×7G8 progeny to 11 showed two groups of recombinants that were separated by a large difference in IC50 as well as some variation in IC50 within each group, particularly in the resistant group (Fig. 4b). QTL analysis identified a major peak on chromosome 4 with an LOD score of 15.6 and some minor peaks on chromosomes 6 and 11 with LOD scores lower than 1.5 (Fig. 4c). A perfect genotype/phenotype association was found at marker C4M69 within a segment of chromosome 4 bounded by crossovers at marker C4M19 in progeny DAN and at marker C3M35 in progeny AUD and JH6 (Fig. 4b). The chromosome 4 segment spanned ~59 kb DNA and contained ten genes (Supplementary Table 9), including the gene encoding P. falciparum dihydrofolate reductase (PfDHFR).

Figure 4

Identification of genetic loci linked to response to trimethoprim (11). (a) Chemical structure of 11. (b) IC50 values (mean and standard deviation) of 11 and allelic designations for three microsatellite markers on chromosome 4 are shown for the parents and progeny of a GB4×7G8 cross. (c) Peaks of quantitative trait loci analysis linked to differential responses to 11 are shown. Predicted genes within the chromosome 4 locus can be found in Supplementary Table 9.

12 (Supplementary Fig. 2a) is a Na+ channel blocker15 with activity against edema from heart, kidney, or liver disease. In assays with 12, the 7G8 parasite had an IC50 at least 70 times higher than that of GB4 (Supplementary Fig. 2b). Thirty-two progeny from the 7G8 and GB4 cross were tested with 12, and two groups were identified that were separated by large differences in IC50 with some variations in IC50 within each group (Supplementary Fig. 2b), suggesting contributions from one major and several minor genes. QTL analysis of the genotypes and phenotypes from the progeny mapped a major determinant to a locus on chromosome 4 with a LOD score of 14.1 and another locus on chromosome 12 with a LOD score of ~2.3. Like 11, the response to 12 mapped to a ~59-kb locus defined by informative crossovers in the same three progeny between marker C4M19 (DAN) and marker C3M35 (AUD and JH6) (Supplementary Fig. 2b), containing pfdhfr and nine other genes (Supplementary Table 9). These results suggested that a single locus confers resistance to both 11 and 12.

Mutations in PfPgh-1 confer resistances to DHMS

Among the genes in the chromosome 5 locus, pfmdr1 was a primary candidate because it encodes an ABC transporter involved in drug resistance, and there are no known targets to 1, such as 5-HT serotonin receptor or α-adrenergic receptor, in the locus12,13,16. To investigate whether pfmdr1 mediated differential susceptibility to 1, we examined pfmdr1 mutations and responses to 1 in parasite isolates from different regions of the world (Supplementary Table 10). There are five known nucleotide substitutions in the pfmdr1 genes, resulting in codon polymorphisms for N86Y, Y184F, S1034C, N1042D, and D1246Y17,18. Examination of the pfmdr1 haplotypes and 1 responses in the field isolates suggested that the N1042D substitution was associated with a greater than 50% reduction in IC50 and the S1034C substitution was associated with a further reduction of 50% or more (Supplementary Table 10). These mutations account for the majority of the difference between 7G8 (the New World pfmdr1 haplotype) and GB4 (the Old World haplotype). The roles of N86Y, Y184F, and D1246Y changes on sensitivity to 1 were unclear, as parasites with the same substitutions had quite different IC50 values in some cases (Supplementary Table 10). Field isolates usually display substantial genetic heterogeneity and the differences in genetic background could contribute to the variations in IC50 values observed.

To evaluate the influence of genetic background on the IC50 measurements, we tested several parasite lines where the pfmdr1 coding regions were replaced with versions carrying different polymorphisms by allelic exchange18,19. Parasites with substitutions on the background of D10, which has the same pfmdr1 allele as GB4, and on the background of 7G8 have been reported 18. Five parasites with different pfmdr1 alleles on 7G8 or D10 genetic backgrounds were compared (Supplementary Table 10). Measurements of IC50 values from three of the five parasites were consistent with the predicted drug-response phenotypes, whereas two parasites (D107G8/1 and D107G8/2) yielded discrepant results. We therefore designed PCR primers to amplify specific sequences from the two parasites (Supplementary Fig. 3). Sequencing of the PCR products showed that the pfmdr1 alleles in the two parasites were not the reported sequences; instead, the amino acids deduced from the DNA sequences matched the IC50 values observed in our drug tests (NYCDY for D107G8/1 and NYSND for D107G8/2; Supplementary Table 10). The apparent discrepancies between genotype and phenotype may have arisen from the mislabeling of parasites or cross-contamination during in vitro culture.

Additionally, we tested six transfected lines with pfmdr1 alleles exchanged in two progeny (GC03 and 3BA6) of a Dd2×HB3 cross19. 1 responses of these allelic-exchanged parasites showed that an amino acid change of N1042D reduced the 1 IC50 by more than half and changes of S1034C and D1246Y lowered it further to less than 300 nM (Supplementary Table 10). Although the F184Y change slightly reduced the IC50 in D10 and 3D7 parasites (compared with C2A and 224), the same change in 7G8D10 and D10D10 had little effect, suggesting that the F184Y substitution had little influence on the response to 1. Together, the results from field isolates and allelic exchanged parasite lines demonstrate a strong association of S1034C and N1042D substitutions and resistance to 1. Of interest, verapamil, a compound that inhibits human P-glycoprotein and can reverse the effects of many antimalarial and anticancer drugs19, did not have any effect on parasite response to 1 (Supplementary Table 11).

Polymorphisms in PfDHFR confer resistance to 11 and 12

Variations in PfDHFR among the seven parasite lines support the association of differential parasite responses to 11 or 12. Although the locus on chromosome 4 contained ten genes, we considered pfdhfr as the primary candidate for resistance to both 11 and 12, because 11 is a known DHFR inhibitor14 and inhibition of human leukocyte DHFR activity by 12 has also been reported20. Many mutations in PfDHFR have been shown to contribute to parasite responses to antifolate drugs, particularly substitutions at amino acid positions at 51, 59, and 10821.Sequencing of the pfdhfr coding region showed only two codon changes between 7G8 and GB4, with GB4 encoding N51 and S108 and 7G8 encoding I51 and T108 (Supplementary Table 12). We also determined seven known polymorphic sites of PfDHFR at codons 16, 50, 51, 59, 82, 108, 140, and 164 from the five other parasites, but found alterations only at codons 51, 59, and 108 (Supplementary Table 12). Similar to the pattern of response to pyrimethamine, a DHFR antagonist, parasites with substitutions at three positions in the gene had higher IC50 values to either 11 or 12 than those with two substitutions, which in turn were more resistant to parasites having one substitution22. These results suggested that mutations in PfDHFR confer resistance to 11 and 12.

To rule out the influence of genetic backgrounds on parasite response, we tested two 3D7 lines that were transformed to express integrated alleles of PfDHFR with S108T substitution from HB3 or triple N51I, C59R, and S108T substitutions from Dd2 (pDT-HB3 and pDT-Dd2; Supplementary Table 12)23. As expected for parasites with integrated alleles, DNA sequencing confirmed the PfDHFR genotypes in the two parasites, which revealed S/T mixed codons at position 108 for pDT-HB3, and mixed codons at positions 51, 59, and 108 for pDT-Dd2 (Supplementary Fig. 4 and Supplementary Table 12). The 3D7 parasite transfected with the HB3 PfDHFR allele had IC50 values similar to those of HB3, while the parasite transfected with Dd2 allele had even higher IC50 values, although the IC50 values were lower than that of Dd2 parasite. These results demonstrated that introduction of a mutant PfDHFR allele into a sensitive parasite increased the IC50 values to both drugs. Sequencing of DNA from progeny of the GB4×7G8 cross showed that pfdhfr alleles in these parasites also matched the phenotypes observed (data not shown). These results, obtained from parasite lines, transfected parasites, and progeny, demonstrate that polymorphisms in PfDHFR indeed confer differential responses to 11 and 12.

DISCUSSION

Progeny from genetic crosses, high-density MS and SNP maps, and high-throughput genotyping methods4,24,25 provide a powerful platform to identify parasite genes involved in differential responses to bioactive compounds. As a proof of principle, we identified a large number of phenotypic variations (DCPs) between individual parasites and linked two genes to parasite responses to three compounds using genetic mapping. Identification of many DCPs is consistent with a highly polymorphic P. falciparum genome26–28. It may also be possible to map DCPs by genome-wide association studies using genetically heterogeneous parasite populations4. The DCPs identified here may lead to the identification of many target genes and to a better understanding of the molecular interactions within the parasite.

Several large-scale screens for antimalarial compounds have been described, most of which assayed compounds at one or two concentrations (typically 10 μM)11,29–33. In contrast, we measured parasite responses to each compound at concentrations across four orders of magnitude, allowing us to generate concentration-response curves and derive IC50 values for all tested compounds immediately (Supplementary Table 2). Many active compounds showed IC50 values less than 1 μM (Supplementary Table 6), some of which may be tested to treat malaria without extensive clinical trials because many of the compounds in the LOPAC collection have been used for treating human diseases. Our study also showed that a compound active against one parasite line was often not effective in killing other parasites with different genotypes.

The observation of more DCPs among parasites resistant to 4 suggests that mutations and compensatory changes selected by 4 may impact parasite responses to other chemicals. Similarly, Dd2 was derived from W2 after selection to 334, and Dd2 was more less sensitive to many compounds than W2 (Supplementary Table 3). These observations suggest that following drug selection, parasites may become more resistant to multiple antimalarial drugs more quickly, consistent with the proposal of parasites showing the accelerated resistance to multiple drugs (ARMD) phenotype35. Because Dd2 and W2 have a very similar genetic background but are known to differ in pfmdr1 copy number, the variation in pfmdr1copy number may play a role in some differential responses.

Mutation and amplification of pfmdr1 play a role in drug responses in P. falciparum18,36,37. There are some similarities in the associations of pfmdr1 alleles to responses to 1 and antimalarial agents such as 2, 3, and halofantrine (13). Amino acid changes in PfPgh-1 (from NFCDY to NFSND) increase the IC50 values of 2, 3, and 13 by two to six fold and decrease the IC50 of 5 in some genetic backgrounds18,19. Likewise, 1 showed a higher IC50 in parasites containing the NFSND allele. However, the changes in IC50 to 2, 3, and 13 after allelic exchange might be parasite specific because 7G8 and GB4 have very similar IC50 values to both 2 and 3 (Supplementary Table 1). Because PfPgh-1 is an ABC transporter homolog, mutations in this protein may affect its ability to transport 1 and other drugs.

Although 12 is a known Na+ channel blocker15, our results showing that 12 acts on PfDHFR were consistent with observations that 12 interferes with folate metabolism in vitro and in vivo20,38,39. Interestingly, the known chemical structures of DHFR inhibitors and epithelial sodium channel blockers share a similar amino-pyrimidine ring and tethered aryl group (Supplementary Fig. 5). It is possible that these structural features account for the inhibition of DHFR and some sodium channels and these proteins may share similar binding pockets40.

Our results showing that 1 interacts with or is transported by an ABC transporter instead of the expected 5-HT2B receptor highlights how such studies can identify “off-target” effects of bioactive molecules. The observation that mutations in an ABC transporter can alter sensitivity to 1 may provide important information for studies and treatment of migraine syndrome.

METHODS

Parasites and compounds

The P. falciparum lines used in this study have been described previously41, and their origins and responses to 2–5 are summarized in Supplementary Table 1. Parasites Dd2, HB3, GB4, 7G8, and 3D7 are parents of three separate genetic crosses7–9. Dd2 is a clone derived from W2 that was originally isolated from Southeast Asia34. Progeny of a genetic cross (GB4×7G8) were described recently9. Transgenic parasites with the gene encoding a homolog of the human P-glycoprotein (PfPgh-1) replaced by mutant alleles were reported previously18,19. The parasites described by Reed, et al.18 were obtained from MR4 (http://www.mr4.org). Parasites transfected with mutant genes encoding PfDHFR were described previously23. All parasites were cultured in vitro according to methods described42. Briefly, parasites were maintained in RPMI 1640 medium containing 5% human O+ erythrocytes (5% hematocrit), 0.5% Albumax (GIBCO), 24 mM sodium bicarbonate, and 10 μg/ml gentamycin at 37°C with 5% CO2, 5% O2, and 90% N2.

The LOPAC1280 collection (http://www.sigmaaldrich.com/chemistry/drug-discovery/validation-libraries/lopac1280-navigator.html) was purchased from Sigma-Aldrich and the compounds prepared and plated as described10. 1–5, 11, and 12 were purchased from Sigma-Aldrich and had purities of 98% or better.

In vitro drug assays

The SYBR Green viability assay (Supplementary Table 13) was adapted from methods described previously11,32. Briefly, 3 μl culture medium was dispensed into 1536-well black clear-bottom plates (Aurora Biotechnologies) using a Multidrop Combi (Thermo Fisher Scientific Inc.); 23-nL compounds in DMSO were added by a pin tool (Kalypsys), and 5 μl of P. falciparum-infected RBCs (0.3% parasitemia, 2.5% hematocrit final concentration) were added. The plates were incubated at 37°C in a humidified incubator in 5% CO2 for 72 h, and 2 μl lysis buffer (20 mM Tris-HCl, 10 mM EDTA, 0.16% saponin, 1.6% triton-X, 10X SYBR Green I supplied as 10,000X final concentration by Invitrogen) was added to each well. The plates were mixed for 25 sec with gentle shaking and incubated overnight at room temperature in the dark. The following morning, fluorescence intensity at 485(14) nm excitation and 535(25) nm emission wavelengths was measured on an EnVision (Perkin Elmer) plate reader. The LOPAC1280 collection (Sigma Aldrich) was screened against each line at eight (seven for Dd2) five-fold dilutions beginning at 29 μM. Antimalarial drugs, 2 and 3, and DMSO were included as positive and negative controls for each plate, respectively.

Follow-up SYBR Green assays in 96-well plate format were performed as described43. Briefly, 150 μl synchronized parasites diluted to 1% parasitemia with 1% hematocrit were mixed with 50 μl medium containing compound. Compound stocks (10 mM) were dissolved in ethanol or DMSO and tested at 11 two-fold dilutions in triplicate. The beginning concentration of each compound was adjusted depending on IC50 values from the initial qHTS. The plates were incubated at 37°C under 5% CO2, 5% O2, and 90% N2 for 72 h, incubated for another 30 min after addition of 50 μl lysis buffer, and read in a FLUOstar OPTIMA reader (BMG Labtech). Data were analyzed using software at the NIAID Bioinformatic Resource website (https://niaid-biocluster.niaid.nih.gov/cgi-bin/bipod2/index.cgi).

Data analysis

Analysis of compound concentration-response data from the qHTS was performed as previously described10. Plate reads were first normalized relative to the control inhibitor (0.29 μM 2) and vehicle (DMSO) wells present on each plate and then corrected by an algorithm using vehicle-only control plates at the beginning and end of the compound plate stack. Concentration-response data for each compound were curve-fitted and categorized into four major classes as described10 using custom software (http://ncgc.nih.gov/pub/openhts/curvefit/). Briefly, compounds in Class 1 produce a complete response curve containing two asymptotes, while compounds in Class 2 have incomplete response curves containing one asymptote. Compounds in Class 1 and 2 curves are further divided based on efficacy, with subclasses 1.1 and 2.1 having 80% or greater efficacy and subclasses 1.2 and 2.2 having less than 80% efficacy. Compounds with Class 3 curves show activity only at the highest concentration or are poorly fit Class 1 or 2 curves with r2 less than 0.9. Compounds with Class 4 curves are inactive and either do not have a curve fit or the curve fit is below three standard deviations of the mean basal activity (about 30% efficacy).

Compound activities were categorized as Active (Class 1.1, 1.2, or 2.1), Inconclusive (Class 2.2 or 3) or Inactive (Class 4). Consensus activity between two replicates was scored as Active (Class 1.1, 1.2 or 2.1 in both replicates or Class 1.1, 1.2 or 2.1 in one replicate and Class 2.2 or 3 in the other), Inconclusive (Class 2.2 or 3 in both replicates), and Inactive (Class 4 in one or both replicates). For determining the potencies of consensus actives (Supplemental Tables 4–6), the IC50 values of both replicates were averaged. Differentially active compounds were defined as having five-fold or greater IC50 diferences between pairs of parasite lines with IC50 values determined as follows: If active in both replicates or active in one replicate and inconclusive in the other replicate, curve fits with P values 0.05 or less were used or averaged. If inconclusive with curve fit P values 0.05 or less for both replicates, IC50 values were averaged. If inactive in both replicates, the IC50 value was set to 29 μM, the highest tested concentration. If inconclusive with a curve fit P value greater than 0.05 or inactive in one replicate, the IC50 value was not used. Unsupervised hierarchical clustering of compound consensus activities and parasite lines was performed using Spotfire DecisionSite (TIBCO Software Inc.).

DNA sequencing, genotyping, and quantitative trait loci (QTL) analysis

To verify nucleotide substitutions in the genes encoding PfPgh-1 and PfDHFR, primers specific to the genes were commercially synthesized to amplify DNA sequences with known polymorphic sites (Supplementary Fig. 3). PCR products were sequenced directly without cloning into a plasmid vector as described44. MS genotypes for the parents and progeny of the genetic GB4×7G8 cross have been reported9,24. QTL analysis was carried out using the J/qtl program45.