Mapping the genomic landscape of multidrug resistance in Plasmodium falciparum and its impact on parasite fitness

Drug-resistant Plasmodium falciparum parasites have swept across Southeast Asia and now threaten Africa. By implementing a P. falciparum genetic cross using humanized mice, we report the identification of key determinants of resistance to artemisinin (ART) and piperaquine (PPQ) in the dominant Asian KEL1/PLA1 lineage. We mapped k13 as the central mediator of ART resistance in vitro and identified secondary markers. Applying bulk segregant analysis, quantitative trait loci mapping using 34 recombinant haplotypes, and gene editing, our data reveal an epistatic interaction between mutant PfCRT and multicopy plasmepsins 2/3 in mediating high-grade PPQ resistance. Susceptibility and parasite fitness assays implicate PPQ as a driver of selection for KEL1/PLA1 parasites. Mutant PfCRT enhanced susceptibility to lumefantrine, the first-line partner drug in Africa, highlighting a potential benefit of opposing selective pressures with this drug and PPQ. We also identified that the ABCI3 transporter can operate in concert with PfCRT and plasmepsins 2/3 in mediating multigenic resistance to antimalarial agents.


INTRODUCTION
Malaria caused an estimated 247 million cases and 619,000 deaths in 2021, mostly in sub-Saharan Africa (1).Efforts to reduce the impact of malaria have repeatedly been stymied by the emergence and spread of Plasmodium falciparum resistance to antimalarial drugs (2,3).Resistance to artemisinin (ART) derivatives, the fast-acting backbone of ART-based combination therapies (ACTs), was first reported in western Cambodia in the late 2000s and has since spread across the Greater Mekong subregion (GMS) (4,5).In patients, resistance manifests as delayed parasite clearance following ART treatment (6).Resistance to piperaquine (PPQ), the long-lasting partner drug of the widely used first-line combination treatment dihydroartemisinin (DHA) + PPQ, was also first detected in western Cambodia and resulted in rapid loss of treatment efficacy across the GMS, with failure rates as high as 87% in northern Thailand (7,8).
Investigations into PPQ resistance initially identified multicopy plasmepsins 2 and 3 (pm2/3) as biomarkers (23,24).These aspartic proteases contribute to hemoglobin proteolysis inside the parasite's acidic digestive vacuole (DV), thereby releasing globin (a source of amino acids) and toxic heme.Subsequent studies of PPQ resistance identified a critical role for mutations (such as M343L or F145I) in the DV membrane-resident chloroquine resistance transporter (PfCRT).These single-point mutations emerged in the GMS on the background of the Southeast Asian mutant PfCRT Dd2 isoform [Fig. 1, legend; (25)(26)(27)(28)], which is known to mediate resistance to CQ, a 4-aminoquinoline that is chemically related to the bis-quinoline PPQ.This isoform also partially reduces parasite susceptibility to amodiaquine while increasing susceptibility to lumefantrine (LMF), both widely used ACT partner drugs (18,(29)(30)(31).PfCRT variants (including M343L, F145I, or other mutations) are thought to mediate PPQ resistance by enabling drug efflux through the central cavity of PfCRT (32,33).This efflux is thought to remove PPQ from the DV, where it normally accumulates as a protonated species and inhibits heme detoxification and hemozoin formation (34)(35)(36).PPQ-resistant parasites display an unusual dose-response profile in which the IC 50 (i.e., the concentration that yields halfmaximal growth inhibition) shows minimal differences relative to PPQ-sensitive parasites, yet at high PPQ concentrations parasites show a biphasic survival curve (37).Defining the relative contributions of PfCRT and pm2/3, in terms of resistance and fitness costs, has until now been elusive.
P. falciparum genetic crosses using simian models have previously proven effective in identifying drug resistance determinants, including pfcrt and antifolate resistance mediators (38).Recent studies have explored ART or CQ resistance using the FRG-NOD humanized mouse model that permits pre-erythrocytic liver stage development and infection of human red blood cells (RBCs) by P. falciparum parasites (39)(40)(41)(42)(43).To further investigate the genetic basis of P. falciparum resistance to DHA and PPQ, we conducted a genetic cross between a Cambodian multidrug-resistant parasite and a drug-sensitive African parasite.Our data uncover secondary markers of ART resistance and elucidate the epistatic relationship between pfcrt and pm2/3 in modulating PPQ resistance.We also identify the ABC transporter (abci3) as an additional member of this gene interaction network involved in differential susceptibility to other antimalarial compounds.Finally, we identify PPQ-resistant pfcrt as a driver of enhanced sensitivity to LMF and mefloquine (MFQ), supporting their combined use in the field.

Implementation of a RF7 × NF54 genetic cross to map determinants of drug resistance
To identify determinants of P. falciparum resistance to DHA and PPQ, we conducted a genetic cross between the PPQ-and DHA-resistant contemporary Cambodian clinical isolate RF7 and the sensitive African parasite line NF54 (Fig. 1).RF7 was selected as representative of the multidrug-resistant KEL1/PLA1/PfPailin lineage that has dominated the GMS (5,10).RF7 expresses the K13 C580Y variant as well as the PfCRT Dd2 + M343L isoform and multicopy pm2/3, with a prior study confirming pm2/3 overexpression (27).We first characterized RF7 and NF54 susceptibility profiles to DHA, PPQ, and a panel of other antimalarials (Fig. 1A).Using the ring-stage survival assay (RSA) to measure ART resistance in vitro, we observed a high percentage of surviving DHA-treated parasites (mean ± SE, 11.8 ± 1.0%), whereas NF54 showed marginal survival (0.2 ± 0.2%).Likewise, using the PPQ survival assay (PSA), ring-stage RF7 parasites gave a high survival rate Fig. 2. Bulk segregant analyses of progeny pools pressured with PPQ, mdCQ, or PYM.The significant genetic loci enriched by each drug in individual drug-drug pairwise comparisons are shown with black lines representing G 0 values (left), along with their corresponding RF7 parental allele frequencies for the set of 18k SNPs that differ between the two parents (right).The G 0 is the smoothed statistical value calculated for individual SNPs and used for detecting QTLs.Regions with a false discovery rate q value < 0.01 (mdCQ versus PYM), < 0.4 (PPQ versus PYM), or < 0.2 (PPQ versus mdCQ) were considered statistically significant QTLs (table S5).Genes enriched by each drug and the RF7 parental allele frequency for the drug-treated samples are colored as follows: mdCQ (orange), PYM (purple), and PPQ (green).Individual allele frequencies for each drug-treated pool are indicated by light dots in the background.The solid colored lines represent the averaged RF7 allele frequency across windows of 100 consecutive SNPs.
RF7 was also highly resistant to monodesethyl chloroquine (mdCQ; the active metabolite of CQ) and pyrimethamine (PYM).In contrast, RF7 exhibited three-to sixfold increased sensitivity to the ACT partner drugs, MFQ and LMF, compared to NF54 (Fig. 1B and table S1).There were no significant differences in IC 50 values for other antimalarials tested, including amodiaquine, pyronaridine, quinine and atovaquone.
To achieve a genetic cross, RF7 and NF54 gametocytes were fed as 1:1 pools to female Anopheles stephensi mosquitoes to initiate sexual stage recombination.We then infected four human liver-chimeric FRG-NOD mice with sporozoites, either via intravenous inoculation of manually dissected sporozoites or via mosquito bites (Fig. 1C and table S2).Haploid ABS parasites were recovered on day 7.5 postinfection and subsequently cultured in vitro (Fig. 1C).

Bulk segregant analyses identify major determinants of P. falciparum resistance to PPQ, CQ, and PYM
Whole-genome sequencing (WGS) of the RF7 and NF54 parents identified 18,489 ("18k") single-nucleotide polymorphisms (SNPs) between their core genomes (tables S3 and S4).These SNPs were located in 2439 P. falciparum coding genes, corresponding to nearly half of the core genome of ~5100 genes, and with an average of ~1 kb distance between SNPs.These SNPs were distributed quite evenly across coding and noncoding regions (fig.S1, A to C).The extensive diversity between these geographically distinct parents enabled high-resolution genetic mapping of drug susceptibility profiles in the cross progeny.
We applied a bulk selection approach to enrich for recombinant ABS progeny resistant to PPQ, mdCQ, or PYM (Fig. 1D).Surviving parasites from these drug-pressured bulk cultures were then subjected to WGS,and parental allele frequencies were measured across the 14 chromosomes (Fig. 1D).Using bulk segregant analysis, we then compared the parental frequencies at each of the 18k SNP positions between drug-treated cultures or between drug-treated and untreated cultures (Fig. 2 and fig.S2).Comparing the PYMtreated bulk against the mdCQ-treated sample or the untreated bulk revealed a significant quantitative trait locus (QTL) on chromosome (chr) 4 (Fig. 2, table S5, and fig.S2A).This locus, spanning 141 kb, contained dihydrofolate reductase (dhfr), which in RF7 parasites encodes a quadruple mutant PYM-resistant allele.By comparison, mdCQ selected for a dominant 856-kb QTL on chr7 that contains the mutant pfcrt Dd2 + M343L allele, as well as two smaller peaks on chr6 and chr12 that contain the putative amino acid transporter aat1 and the v-type pyrophosphatase vp2.mdCQ and PPQ showed similar allelic frequency profiles, consistent with their similar modes of action (35).PPQ, however, selected for a 117-kb chr14 segment containing the tandem pm2/3 amplicon.In contrast, PPQ selected against a 484-kb segment on chr13 harboring mutant E415G exonuclease-I, which had previously been proposed to be a molecular marker of PPQ resistance (Fig. 2 and fig.S2) (24).
Although bulk segregant analysis can identify loci containing the major resistance mediators for these drugs, this approach may be constrained by the fitness cost of certain genotypes that associate with reduced expansion in bulk cultures.To refine the identification of resistance markers and map secondary determinants, as well as understand epistatic interactions between genes, we performed QTL analysis with individual clones (Fig. 1E).

Drug pulses enrich for genetically and phenotypically diverse progeny WGS analysis identifies a set of genetically distinct recombinant progeny
We cloned 148 individual progeny by limiting dilution of untreated cultures and identified 66 recombinant progeny using WGS analysis of the 18k SNP set.The remaining 82 progeny proved to be genetically identical to NF54, an indication of a high selfing rate for this parent in our cross.The 66 recombinant clones that were derived in the absence of drug segregated into 12 unique haplotypes.Eight recombinant haplotypes were recovered from the one mouse that was intravenously inoculated with sporozoites, whereas only six recombinant haplotypes were recovered from the three mice infected by mosquito bites (Fig. 3A and table S2).An additional 81 recombinant clones were obtained from bulk cultures pressured with PPQ, mdCQ, or PYM, yielding an additional 22 unique haplotypes (Fig. 3, A and B).Our analysis focused on these 34 genetically distinct haplotype groups (Fig. 3B), in addition to parental NF54 and RF7 (referred to as HapA and HapN, respectively).Unexpectedly, we observed minimal overlap in the sets of recombinant progeny selected using PPQ and mdCQ pressure.Also, no common haplotypes were obtained using PYM pressure compared with either of the other two drugs (Fig. 3B).PPQ and mdCQ generated five and six unique recombinant haplotypes, respectively, and one additional shared haplotype.This finding suggests that resistance to each drug requires distinct combinations of genetic determinants.

Genetic analysis reveals recombination rates, inheritance patterns, and relatedness among progeny
Our analysis of the 34 unique recombinants revealed a recombination rate of 15.1 kb/cM and a positive correlation (R 2 = 0.47) between the number of crossover events and physical lengths (table S6 and fig.S1, D and E), similar to previous crosses [12.1 to 14.3 kb/cM; (40,44,45)].Most clones shared ~50% similarity in parental allelic composition across the genome, as measured by identity-by-descent analysis (fig.S1F).Nonetheless, using multidimensional scaling of pairwise genetic distances, we observed that the recombinant progeny clustered based on their selection conditions (fig.S1G).Progeny derived with PPQ or mdCQ pressure clustered more closely and were distinct from the progeny pressured with PYM or derived without drug.This observation agrees with our bulk segregant results and aligns with our expectations based on drug chemical relatedness.There was also significantly higher inheritance of RF7 SNPs in the recombinants obtained following PPQ or mdCQ exposure compared to those obtained in the absence of any drug pressure (Fig. 3, B and C, and fig.S1H), suggesting the possibility of multiple determinants existing in linkage disequilibrium.

PPQ and mdCQ show differential selective pressure for the KEL1/PLA1/PfPailin haplotype
We observed that all progeny selected by PPQ were of the KEL1/ PLA1/PfPailin haplotype (k13 C580Y, k13-flanking region, and multicopy pm2/3), identical to RF7 (Fig. 3D, fig.S3A, and table S7).However, only one of the seven mdCQ-selected progeny inherited this haplotype.While mdCQ treatment enriched for progeny clones harboring mutant pfcrt, this drug selected against a segment on chr14 encompassing pm2/3 (Fig. 3B).This observation suggests that amplified pm2/3 may have a fitness cost in the presence of RF7 mutant pfcrt.Likewise, the majority of recombinants obtained under PYM or drug-free conditions did not harbor the KEL1/PLA1/PfPailin lineage (Fig. 3D).This haplotype's significant association with PPQ-selected progeny (P = 0.009, Fisher's exact test), but not with the mdCQ or PYM-selected progeny, suggests that PPQ favors this particular lineage.

Genomic segregation distortions map to mutant pfcrt and a chr14 loci
To identify loci under segregation distortion in the progeny, we aggregated the distinct sets of clones recovered without or with drug selection in silico and examined their allele frequencies.In the drug-negative clones, there was extreme inheritance bias toward NF54 wild-type (WT) SNPs at multiple loci on chromosomes 1, 7, 13, and 14.These skews were reversed by PPQ, mdCQ, or PYM treatments, which enriched for RF7 SNPs in these segments (Fig. 3E).This reversal was most prominent for pfcrt found on chr7 in mdCQ-or PPQ-pressured parasites, implying a strong fitness cost for mutant pfcrt parasites in the absence of selective pressure.Using the "pooled" drug-selected progeny clones, we also identified an additional 251-kb segment on chr14 containing arps10, which was enriched by all three drugs (Fig. 3E).In addition to pfcrt, the PPQselected clones showed 100% inheritance of RF7 SNPs for chr14 (377 kb) and chr13 (516 kb) segments containing pm2/3 and k13, respectively (Fig. 3E).This observation supports our finding that PPQ positively selected for the KEL1/PLA1/PfPailin lineage.In addition, PPQ, but not mdCQ or PYM, selected for a 416-kb chr6 segment encompassing aat1, suggesting a possible involvement of this marker with PPQ resistance.

Drug exposure enriches for progeny with diverse phenotypic responses
To characterize the DHA and PPQ phenotypic responses of the 34 recombinant haplotypes, we profiled a panel of representative progeny and their RF7 × NF54 parents in drug susceptibility assays (Fig. 4 and table S8).We observed a wider range of phenotypic responses and higher survival and AUC values to either DHA or PPQ for the clones derived post drug selection compared to the nondrug pressured group (Fig. 4, C and D).Notably, several of these drug selected progeny clones had even higher DHA RSA survival and AUC values compared to parental RF7.This result was unexpected as the highly DHA-resistant progeny were derived post-PYM selection.For PPQ, five of the six PPQ-selected clones showed high levels of PPQ survival and resistance, yet none had higher PSA survival than the parental RF7.This finding suggests a large fitness cost even among the PPQ-resistant progeny and reinforces the importance of applying drug pressures to diversify genetic cross progeny phenotypes.

DHA resistance associates with k13 and other secondary determinants QTL mapping of DHA resistance reveals k13 as the primary determinant
We next performed linkage analysis to uncover QTLs associated with DHA resistance, using the phenotype-genotype data from the progeny clones (tables S4 and S8).There was a good correlation (R = 0.85) between RSA and AUC values (fig.S4).QTL mapping using either sets of values revealed a dominant peak on chr13.This 183-kb segment contained 109 SNPs including 47 nonsynonymous mutations within 20 genes and included k13, the primary ART resistance mediator (Fig. 5, A to D).We also observed coinheritance of this segment, spanning −133 to +47 kb of the k13 gene, in all mutant progeny (fig.S3).This observation suggests the possible presence of additional loci that have an impact on ART resistance or that compensate for altered mutant K13 function.Grouping the recombinant progeny by their k13 genotypes (C580Y versus WT) showed a clear segregation in DHA response, with mutant progeny showing higher RSA survival and AUC levels (Fig. 5E).Reverting C580Y to WT in RF7 parasites using CRISPR-Cas9-based gene editing resulted in a significant reduction in RSA and AUC levels (from 15.9 to 4.6% and from 126.6 to 56.3, respectively; Fig. 5F).By comparison, NF54 RSA and AUC levels were 0.2% and 43.3, respectively (Fig. 1A and table S8).G and H) Linkage of chr1 (G) and chr14 (H) segments with pfcrt from QTL analysis using the pfcrt genotype as an outcome.Shown are the significant QTLs above the 95% probability threshold for each analysis.(I) LOD plot for AUC levels after adjusting for k13 as a covariate suggests independent inheritance of the chr14 segment with k13 and coinheritance of the chr1 segment with k13.(J) Scatterplot of AUC levels in progeny segregated by k13, mrp1 (chr1), and arps10 (chr14) genotypes, showing that the RF7 chr14 segment is associated with increased AUC levels.We note that none of the progeny harbored MUT mrp1 and WT k13 alleles.Significance between groups was tested using Mann-Whitney U tests.*P < 0.05, ***P < 0.001.MUT, mutant.

QTL mapping uncovers potential secondary determinants on chr1 and chr14
In addition to the chr13 segment, we observed two smaller QTLs on chr1 and chr14 that were significantly associated with DHA resistance (Fig. 5, A, B, G, and H, and table S9).We then examined whether these associations were additive or epistatic in modulating DHA response.Applying k13 as a covariate resulted in the loss of the chr1 peak for both RSA and AUC, suggesting that the chr1 segment was likely coinherited with k13 (Fig. 5I and fig.S5A).This was evident by the lack of any difference in RSA levels when stratified by the k13 and chr1 genotypes (Fig. 5J).In contrast, the chr14 segment showed a significant QTL peak for AUC either when correcting for k13 genotypes or in the subset of k13 mutants (Fig. 5I, fig.S5B, and table S9).Mutant k13 progeny exhibited higher mean AUC levels when harboring RF7 alleles for chr14, suggesting that this segment might enhance parasite survival at higher doses of DHA (Fig. 5J and table S9).
Within the chr1 185-kb segment, we observed nonsynonymous mutations in 13 genes, including the drug-resistant candidates mrp1 and pip5k (table S9).The 245-kb segment on chr14, notable in only the AUC outcome, comprised 14 genes including arps10.Examination of sequence diversity in mrp1, pip5k, and arps10, across the set of ~2500 sequenced P. falciparum genomes (46), suggested geographic differences with the RF7 parental mutations being prevalent in clinical isolates from Cambodia, Mali, and Ghana (fig.S5C).These chr1 and chr14 segments also showed linkage disequilibrium with pfcrt (Fig. 5, G and H, and fig.S5, D and E).

PPQ resistance is conferred by mutant Dd2 + M343L pfcrt and multicopy pm2/3
To identify the markers of PPQ resistance, we performed QTL mapping using the recombinant progeny values obtained from the PSA and AUC data (table S8).Both datasets identified a major QTL on chr7, which mapped to pfcrt (Fig. 6, A to C, and table S10).Our AUC analyses also identified an additional locus on chr14 harboring the conserved pm2/3 amplicon (Fig. 6, B and  C; fig.S6A; and table S10).Controlling for pfcrt as a covariate revealed no other peaks in the PSA analysis, suggesting that the pfcrt RF7 allele is the main determinant of survival at 200 nM PPQ (fig.S6B).In contrast, controlling for pfcrt improved the logarithm of the odds (LOD) score for the pm2/3 segment for AUC.Reciprocally, controlling for pm2/3 revealed a higher LOD score for the pfcrt segment (fig.S6B).Our results suggest a link between pfcrt, pm2/3, and PPQ AUC.

The epistatic interaction between mutant pfcrt and multicopy pm2/3 modulates PPQ resistance levels
Examining the PPQ response in WT pfcrt progeny revealed a small but significant association between pm2/3 amplifications and higher AUC levels (Fig. 6, D and E).Mutant pfcrt progeny harboring multicopy pm2/3 showed a larger increase in both PSA and AUC levels, coinciding with a biphasic response at high PPQ concentrations (fig.S7).These findings suggest an epistatic interaction between pm2/3 and pfcrt in modulating high-grade PPQ resistance.
To further examine the relative contributions of pm2/3 and pfcrt, we took advantage of a spontaneous loss of pm2/3 copy number in the RF7 parental line that occurred over ~3 months of continuous in vitro culture and used limiting dilution to obtain clones with three copies or a single copy of pm2/3.The multicopy pm2/3 clone exhibited significantly higher PSA and AUC levels and displayed a biphasic PPQ dose-response curve (Fig. 6, F and I, and fig.S8).These data confirm that pm2/3 amplifications can augment PPQ resistance in the presence of mutant pfcrt.
We validated these markers by editing pfcrt in progeny that corresponded to two unique haplotypes and differed in their pm2/3 copy numbers.The HapD progeny was chosen as it represented a DHA-resistant, PPQ-sensitive line.Introducing the mutant Dd2 + M343L pfcrt allele into HapD WT pfcrt parasites having a single copy of pm2/3 did not affect PSA levels and led to a minor increase in AUC levels, suggesting that this mutant pfcrt allele did not suffice to confer PPQ resistance in this progeny (Fig. 6, G, H, and J).
We also selected HapU as it had a KEL1/PLA1/PfPailin haplotype, was DHA and PPQ resistant, and exhibited the highest PPQ resistance level amongst our recombinant progeny (tables S7 and S8).As with RF7, this line was observed to spontaneously deamplify pm2/3 and limiting dilution yielded HapU clones with one to three pm2/3 copies.We observed that the HapU mutant Dd2 + M343L pfcrt clones showed step-wise increases in PSA and AUC levels with increasing pm2/3 copy numbers (Fig. 6, G, H, and K).Removing the pfcrt M343L mutation from these parasites yielded a significant reduction in PSA and AUC levels (Fig. 6, G, H, and L).Our data also provided evidence that the amplification of pm2/3 might constitute a first step of resistance as shown by enhanced parasite survival at lower PPQ concentrations (up to 100 nM).In the presence of multicopy pm2/3, the pfcrt M343L mutation further boosted parasite survival upon exposure to higher PPQ concentrations, and generated a biphasic response (Fig. 6, H and K).Our results reveal an epistatic genetic interaction between pfcrt and pm2/3 in modulating levels of PPQ resistance.

Multicopy pm2/3 and mutant k13 associate with reduced fitness in the RF7 KEL1/PLA1/PfPailin parasite
We next investigated the impact of k13 and pm2/3 resistance determinants on parasite fitness in the absence of drug pressure.To study this, we performed long-term pairwise competitive assays between k13 C580Y mutant versus WT clones that were edited in isogenic RF7 clones harboring one or three copies of pm2/3 (Fig. 7A).The k13 C580Y mutation exhibited a small fitness cost in the presence of multicopy pm2/3, whereas this k13 mutation was fitness neutral on the single-copy pm2/3 background (Fig. 7, B and C).A greater fitness cost was observed with multicopy pm2/3 compared to single-copy pm2/3, in k13 C580Y RF7 parasites (Fig. 7, B and C).These data are consistent with our observation of the reduction of pm2/3 copies in RF7 and HapU bulk parasites in the absence of selection pressure (fig.S8).

The Dd2 pfcrt allele increases parasite susceptibility to the ACT partner drugs LMF and MFQ
Further profiling of the genetic cross parents revealed that RF7 has a three-to sixfold higher sensitivity to both LMF and MFQ compared with NF54 (Figs. 1B and 8, A and B).Phenotypic characterization of the progeny revealed a positive correlation between the IC 50 values for both compounds (R = 0.66), indicative of similar although not identical mechanisms of resistance (Fig. 8, A and B, and fig.S4).This contrasted with the negative correlations (in the range of −0.35 to −0.45) observed between LMF or MFQ IC 50 responses and either PPQ IC 50 or AUC levels among the progeny (fig.S4).This observation is consistent with different markers of resistance and/or opposing alleles for the same set of markers.QTL analyses for LMF IC 50 values revealed a dominant peak on chr7 within 11 kb of pfcrt (Fig. 8C and table S11).For MFQ, we observed multiple QTL peaks (including the same chr7 segment), suggesting a multigenic trait (Fig. 8D).When controlling for pm2/3 as a covariate, we observed an increased association between the chr7 pfcrt QTL and MFQ IC 50 values (fig.S9).Analysis of the IC 50 values and pfcrt genotypes of the recombinant progeny revealed an association between mutant pfcrt and lower IC 50 values for both LMF and MFQ (Fig. 8E).
To identify secondary markers of LMF and MFQ susceptibility, we performed a QTL analysis for LMF IC 50 controlling for pfcrt.These data revealed significant QTLs on chr2 for LMF and chr10 for LMF and MFQ (fig.S9A and table S11).These segments contain atg11 and atg18, respectively.Notably, the atg18 T38I mutation (inherited from the RF7 parent) was associated with lower LMF and MFQ IC 50 values in both WT and mutant pfcrt progeny (Fig. 8F).Progeny harboring both mutant atg18 and mutant pfcrt were the most sensitive to these two drugs, whereas the double WT haplotypes were the least sensitive (Fig. 8F).
We next tested whether pfcrt might directly affect LMF and MFQ responses by assaying our pfcrt-edited isogenic progeny (Fig. 8, G  and H).HapD progeny engineered to carry the mutant Dd2 + M343L pfcrt allele showed a significant threefold reduction in LMF and MFQ IC 50 values compared with the WT pfcrt counterpart (Fig. 8, G to J).However, the pfcrt Dd2 + M343L variant showed no change in its LMF and MFQ IC 50 values when compared to the Dd2 isogenic comparator (Fig. 8, G and H), suggesting that the eight-amino acid Dd2 pfcrt variant allele is solely responsible for heightened sensitivity to LMF and MFQ.We note a small but significant decrease in MFQ IC 50 with higher pm2/3 copies in the HapU pfcrt-edited progeny (Fig. 8, H and K).

RF7 and NF54 exhibit differential susceptibility to preclinical compounds
We tested the RF7 and NF54 response to a panel of 15 clinical and preclinical antimalarial compounds to map modulators of parasite susceptibility (Fig. 9A).Three of these compounds showed higher IC 50 values in RF7.These include the antifolate P218 (47), which targets P. falciparum DHFR that in the case of RF7 is a quadruple mutant that mediates high-level resistance to the related antifolate PYM (2).Four-to 10-fold higher IC 50 values were also observed for MMV665939 and MMV675939 (labeled collectively as "MMV6x").In vitro resistance to the MMV6x5939 compounds was previously associated with single point mutations or CNVs in abci3 (48,49).However, none of these resistance-conferring abci3 mutations or CNVs were present in RF7 (fig.S10A).Nonetheless, the two abci3 point mutations S2106C/S2227G in RF7 showed evidence of selection in field isolates (fig.S10B).This double mutant was present in ~37% of Cambodian or Thai isolates (n = 718) based on our analysis of 2512 isolates sequenced from Asia and Africa (50).

QTL mapping identifies pfcrt as a major determinant of a multigenic trait of reduced parasite susceptibility to MMV6x compounds
To map determinants of resistance to these MMV6x compounds, we phenotyped the cross progeny and performed QTL analysis (fig.S11A and Fig. 9B).This analysis showed that a narrow chr7 peak harboring pfcrt was the strongest predictor of MMV665939 IC 50 and MMV675939 AUC levels (Fig. 9, C and D, and table S12).For both compounds, we also noted a smaller secondary QTL on chr3 containing the abci3 locus.No other QTL peaks were observed for MMV665939 when controlling for pfcrt or pm2/3 (fig.S9B).This observation suggests that MMV665939 resistance is associated with chr7 and chr3 segments, likely pfcrt and abci3.For MMV675939, we observed an additional peak on chr 14 that included pm2/3 (Fig. 9D and fig.S9B).These data suggest that these three loci combine to underpin differential MMV675939 susceptibility.

Multicopy pm2/3, abci3, and pfcrt are combinatorial in modulating parasite susceptibility to MMV6x compounds
In our progeny, higher MMV665939 IC 50 values and MMV675939 AUC levels were associated with both mutant pfcrt and abci3 (Fig. 9,  E and F).Higher MMV675939 AUC levels were also observed in progeny harboring multicopy pm2/3 (Fig. 9F).For MMV675939, biphasic responses were observed either in progeny harboring mutant pfcrt or in progeny harboring mutant abci3 in the presence of multicopy pm2/3 (fig.S12).Gene editing of HapD by replacing WT with mutant pfcrt confirmed the role of this gene in MMV6x response (Fig. 9, G and H) and resulted in a biphasic MMV675939 dose response (Fig. 9I).Amplified pm2/3 decreased parasite susceptibility to MMV675939 in HapU and HapU-edited lines, whereas MMV665939 was unaffected (Fig. 9, G and H).Genetically edited HapU parasites showed that the pfcrt M343L mutation had no impact on either MMV6x compound (Fig. 9, G and H).These data lead us to conclude that the Dd2 pfcrt allele itself can cause reduced susceptibility to MMV6x compounds and produce a biphasic response to MMV675939 (Fig. 9, J and K).

MMV6x compounds can interfere with mutant PfCRTmediated efflux of PPQ and CQ
To investigate whether resistance to the two MMV compounds is attributable to PfCRT transport, we performed competitive inhibition assays to test their ability to inhibit PPQ or CQ uptake using proteoliposomes loaded with PPQ-or CQ-resistant mutant PfCRT isoforms, respectively (32).Control assays showed that CQ substantially inhibited PPQ uptake and vice versa.Verapamil, a known CQ resistance reversal agent, partially inhibited CQ and PPQ uptake.These results are consistent with earlier findings that verapamil can reverse resistance to both drugs via competitive inhibition of PfCRT-mediated drug transport (27,36,51).MMV675939 and MMV665939 both competed with PPQ and CQ for mutant PfCRT-mediated drug uptake, suggesting that these compounds can interact directly with PfCRT (fig.S11, B  and C).

DISCUSSION
By leveraging a P. falciparum genetic cross, we identified parasite loci associated with resistance to multiple first-line antimalarial drugs.Our studies of ART resistance identify two unlinked chromosomal segments that appear to contribute to ART resistance, in addition to mutant k13 that our genotype-phenotype mapping localized as the primary determinant.We also report that PPQ resistance is dictated by an epistatic relationship between mutant pfcrt and multicopy pm2/3, which together achieve high-level resistance.However, this level of resistance comes with a fitness cost that leaves them vulnerable to replacement with WT pfcrt or single-copy pm2/3 in the absence of PPQ pressure.We also show that the firstline partner drug LMF and its related arylaminoalcohol MFQ select against mutant pfcrt, underscoring a strategic advantage of implementing PPQ with either LMF or MFQ in the same field settings.Last, our data suggest that mutations in abci3 form a network with pfcrt and multicopy pm2/3 in regulating parasite susceptibility to diverse chemotypes, as illustrated with the two antiplasmodial drug discovery candidates MMV665939 and MMV675939.
Recent reports that k13 mutations have emerged in several sites in east Africa and the Horn of Africa are a cause for concern, as reduced ART efficacy increases selective pressure on the partner drug and raises the prospects of future treatment failures (52).Intravenous artesunate is also first-line treatment for severe malaria (53), with mutant k13-mediated delayed parasite clearance potentially compromising treatment efficacy.Mutant k13 parasites in the GMS are often more ART resistant and display improved fitness compared with African strains, suggesting that Asian strains have acquired secondary determinants (12,22).Our genetic cross data identify two candidate loci-a chr1 segment harboring mrp1 and a chr14 segment harboring arps10.The four mrp1 mutations present in RF7 (H191Y, S437A, I876V, and F1390I) are quite prevalent in the GMS and have been associated with altered susceptibility to ART derivatives (54,55).The most common of these mutations in Africa, I876V, was earlier associated with parasite recrudescence following artemether-LMF treatment (56).This transporter is nonessential in P. falciparum ABS parasites cultured in vitro, with knockout parasites showing increased parasite susceptibility to ART (57).In our genetic cross progeny, mutant mrp1 was only found in parasites harboring mutant k13.
Our arps10 marker on chr14, which harbors the V127M + D128H mutations, has been previously associated with prolonged parasite clearance half-lives following artesunate treatment and with increased RSA survival in K13 mutant parasites (58,59).This gene is thought to be a key founder mutation of the KEL1/ PLA1 lineage on which ART resistance emerged in the GMS (58).Other mutations in this chr14 segment include exonuclease-V R25K and the alpha/beta hydrolase I301F, which were earlier identified as potential markers of DHA-PPQ treatment failures in Cambodian isolates (24).It is noteworthy that this segment overlaps the QTL associated with enhanced fitness from separate genetic cross studies with GMS parasites (60).These genes on chr 1 and 14 merit further examination as potential contributors to ART resistance emerging in African settings.
Our PPQ studies showed unexpected connections to ART resistance.Upon applying PPQ or CQ pressure to bulk progeny, we recovered resistant pfcrt-mutant progeny which were all mutant k13.PPQ also selected for the ART-resistant chr14 segment.These observations suggest that the selection of pfcrt mutants by PPQ in the field might have evolved on the background of mutant k13 and favored the inheritance of the chr14 segment that might augment DHA resistance.Our results provide support that PPQ may have contributed to the dominance of the combined KEL1/PLA1 (k13 C580Y and multicopy pm2/3) colineage across the GMS where DHA-PPQ was commonly used (28).The interconnection between PPQ and ART resistance presumably centers on heme, which serves as a target for PPQ and an activator for ART (35).
Our PPQ studies resolve debate about the relative roles of mutant pfcrt and multicopy pm2/3 in driving resistance to this drug (fig.S13) (27,28,37).At PPQ concentrations <100 nM, we observed a decrease in parasite susceptibility upon amplification of pm2/3 in isogenic progeny expressing either mutant Dd2 + M343L or Dd2 PfCRT isoforms.At 200 nM PPQ or above, the Dd2 + M343L variant afforded a clear gain of PPQ resistance, which increased with increasing pm2/3 copy number.The biphasic nature of the PPQ dose-response curve was most pronounced with multicopy pm2/3 on a mutant Dd2 + M343L pfcrt background.
The PPQ C max in plasma is ~200 nM (61), which, based on our quantitative phenotypic analysis, would be consistent with both determinants contributing to resistance.The long terminal half-life of PPQ (~3 weeks) would also explain a selection advantage for multicopy pm2/3, which we found to enhance parasite survival at low PPQ concentrations.This partial gain of resistance likely favored the emergence of novel PfCRT variants that dominate PPQ resistance at higher drug concentrations (62).Biophysical studies using proteoliposomes attribute this to a mutant PfCRT-mediated gain of PPQ efflux away from its heme target in the DV (32).PfCRT-mediated PPQ or CQ efflux has also been shown using heterologous Xenopus oocyte or yeast assays (33,36,63,64).
PPQ and CQ appear to occupy partially overlapping sites in PfCRT, with some mutations including M343L contributing to PPQ resistance while becoming less CQ resistant (27,32,36).Notably, there is no established method to directly measure the impact of PfCRT isoforms on drug transport across the DV in parasites cultured in vitro, as was achieved earlier with PfMDR1 using fluorogenic substrates (65).For pm2/3, the basis for resistance might relate to its impact on the rate of hemoglobin proteolysis or conversion of heme to hemozoin, effectively depleting levels of heme that act as a ligand for PPQ binding.The earlier observation that pm2 and pm3 form part of a multicomponent hemozoin formation complex (66) may explain why amplification of both tandem genes is associated with PPQ resistance, yet overexpression of each gene individually did not affect the PPQ dose-response profile (67).
Our studies revealed that multicopy pm2/3 exerts a substantial fitness cost in the Cambodian RF7 k13 C580Y parasite, exceeding that observed with the k13 mutation itself.We also observed spontaneous deamplification of pm2/3 in parasites cultured for ~3 months in the absence of PPQ pressure.PPQ resistance-associated mutant PfCRT haplotypes are also known to exert a fitness cost (27,68,69).This growth defect has been attributed to an excessive buildup in the DV of globin-derived short peptides that serve as a natural substrate for PfCRT, thereby restricting protein synthesis (70)(71)(72).These observations help explain why the prevalence of both multicopy pm2/3 and PPQ-resistant mutant pfcrt decreased rapidly following withdrawal of DHA-PPQ from first-line use in Cambodia (62).
A recent field-based and genetic cross study provided evidence for coselection of mutations in PfCRT and the putative DV-resident amino acid transporter PfAAT1 (43).RF7 harbors the S258L plus F313S PfAAT1 variant that was selected under CQ pressure in this previous cross (between NF54 and NHP4026).Our bulk segregant analysis of CQ-pressured bulk progeny pools also showed evidence of a minor peak at the pfaat1 locus, in addition to strong selection for mutant pfcrt.Similar QTL studies with our progeny clones did not identify pfaat1 under either CQ or no drug selection.We did, however, observe a peak for this segment in PPQ-pressured progeny.Further studies are clearly warranted to define the role of pfaat1 in P. falciparum resistance to PPQ and CQ across different genetic backgrounds.
In some regions of the GMS, DHA-PPQ has been replaced by artemether-LMF, which to date has remained effective as LMF has not encountered resistance.Artesunate-MFQ has also been reintroduced into the GMS.Our QTL analysis data show a strong association between LMF or MFQ IC 50 values and pfcrt, with these drugs selecting against the Southeast Asian mutant Dd2 + M343L allele.We confirmed these results in pfcrt gene-edited parasites.These data provide evidence that LMF and PPQ can exert opposing selective pressures on the PfCRT locus, in agreement with earlier observations that LMF selects against CQ-resistant pfcrt as detected using the K76T marker (31,73).
In Africa, artemether-LMF accounts for 75% of the antimalarial drug market (1).Recent studies have shown excellent efficacy of DHA-PPQ either for intermittent preventive treatment of infants or pregnant women or for seasonal malaria chemoprevention (74).Our data lend support to the introduction of DHA-PPQ into areas with artemether-LMF use.Clinical teams have also documented excellent clinical efficacy of triple therapies that combine DHA-PPQ with MFQ in the GMS, with the rationale that use of these triple ACTs can suppress the emergence of multidrug resistance (75).We observed a slight sensitization of our multicopy pm2/3 progeny toward MFQ, a drug that is known to select for multicopy pfmdr1.However, no association was observed with LMF.These data may help explain why P. falciparum parasites in the GMS very rarely show amplification of both pfmdr1 and pm2/3 (62).
Our genetic mapping of candidate loci is constrained by the choice of resistant parent and the resolution of the meiotic crossover events.Nonetheless, our data illustrate the power of genetic crosses for identifying polygenic networks of P. falciparum multidrug resistance.We note that two separate genetic crosses in the humanized mouse model were recently reported with Southeast Asian parasites harboring the ART-resistant k13 C580Y mutation (41).Bulk segregant analysis of DHA-treated synchronized progeny pools only identified the k13 segment on chr13.In our study, we profiled the DHA response (using RSA and AUC) for each recombinant progeny and parent and uncovered distinct peaks in addition to k13.For PPQ, our QTL analysis on individual progeny identified pm2/3 as a second peak that was not observed in the bulk segregant analysis of PPQ versus untreated bulk pools.These data provide evidence that bulk segregant analysis is effective at identifying primary determinants, but QTL analyses of individual progeny can provide more granular resolution of complex genetic traits and uncover epistatic interactions.
Crosses provide an important complement to genome-wide association studies as well as in vitro evolution and WGS studies that can identify candidate chromosomal segments and specific loci (23,24,48,76).These approaches will be key to identifying the genetic basis of P. falciparum resistance to first-line combination therapies, which appears to be imminent in Africa, and leveraging that knowledge to map the spread of resistance and develop strategies to mitigate its impact.

P. falciparum infections in A. stephensi mosquitoes and humanized chimeric liver mice Parasite lines, gametocyte cultures, and mosquito feeding
The RF7 [Sanger ID: PH1008-C ( 27)] Cambodian isolate was provided by R. Fairhurst and was a recrudescent sample of PH1356-C (genetically identical except that the latter had only a single copy of pm2/3).The NF54 isolate was provided by P. Sinnis.RF7 and NF54 gametocytes were cultured as previously described (77).Briefly, parasites were grown in RPMI 1640 containing 10% (v/v) human serum and 4% red blood cell (RBC) hematocrit at 37°C in a glass candle jar.Cultures were initiated at 0.5% asynchronous asexual parasitemia from a low-passage stock and maintained up to day 18 with daily media changes but without any addition of fresh erythrocytes.Day 15 to 18 cultures, containing largely mature gametocytes, were used for mosquito feeds.Cultures were centrifuged (108g, 4 min), and the parasite pellet was resuspended to 0.6% gametocytemia in a mixture of human O + RBCs supplemented with 50% (v/v) human serum.To genetically cross RF7 and NF54, we used a 1:1 ratio of gametocytes at a final gametocytemia of 0.6%.Gametocytes were fed to 2-to 4-day-old A. stephensi mosquitoes (Liston strain) that had been starved overnight, using glass membrane feeders.Unfed mosquitoes were removed after feeding.Fed mosquitoes were then maintained on a 10% sugar solution at 25°C and 80% humidity with a 14-hour:10-hour (light:dark) cycle including a 2-hour dawn/dusk transition.Human RBCs were collected weekly from healthy volunteers, with informed consent, under the Johns Hopkins University Institutional Review Board approved protocol NA_00019050.All experiments were performed in accordance with institutional guidelines and regulations.

Oocyst and salivary gland sporozoite quantification
On day 7 postinfected blood meal, mosquito midguts were harvested and stained with mercurochrome.Oocysts were counted by brightfield microscopy using a Nikon E600 microscope with a PlanApo 10× objective.We observed a 75% prevalence of infection, with an average of 10.7 oocysts per mosquito.Between days 14 and 16, salivary glands were harvested from a minimum of 20 mosquitoes and homogenized, and the sporozoites were counted using a hemocytometer.On average, we obtained 25,000 salivary gland sporozoites per mosquito.

Mouse infections
All animal experiments were performed in accordance with the Johns Hopkins University Animal Care and Use Committee guidelines (approved protocol M017H325).HuHep mice were purchased from Yecuris (FRG-NOD, human hepatocyte repopulated 70%+; catalog no.10-0013).Upon arrival, we began withdrawal of the mice from 2-(2-nitro-4-trifluoromethylbenzoyl)-1,3-cyclohexanedione.Mice were kept in a sterile facility with sterile bedding, food, and water.Mice were weighed daily, and those that showed any weight loss >10% (compared to preshipment weight) were treated with 300 μl of sterile saline via intraperitoneal injection and given a nutritional supplement in water (STAT; https:// prnpharmacal.com/products/nutritional-supplements/stat/).Mice were allowed to recover from shipping for 2 to 3 days before infection.Infection with P. falciparum sporozoites was performed either by mosquito bite or by intravenous inoculation.For the mosquito bite approach, mosquitoes were allowed to probe/feed on the isoflurane-anesthetized mice for 15 to 20 min.Each of the three mice received approximately 50 to 60 infectious mosquito bites.A fourth mouse was intravenous-inoculated with 3 million sporozoites dissected from mosquito salivary glands.Both routes were chosen on the basis of earlier reports of successfully using either intravenous injection or mosquito bites (39,40).Mice infected by mosquito bite remained stable with a healthy appearance for the remainder of the experiment.In contrast, the intravenous-inoculated mouse lost >10% body weight and on day 4 post-inoculation appeared ill.This latter mouse was given 3000 U of penicillin via the intramuscular route, 1000 U of penicillin and 1 mg of streptomycin intraperitoneally, and additional daily doses of penicillin and streptomycin until sacrifice.On days 6 and 7 post-sporozoite infection, 500 μl of RPMI-washed human RBCs was inoculated intravenously into each mouse.At day 7.5 postinfection, ABS parasites were detected at up to 0.3% parasitemia, and blood was recovered by cardiac puncture of mice anesthetized with ketamine/xylazine.One mouse, M46056, which received 3 million sporozoites via the intravenous route, showed the highest parasitemia on day 7 and was used for bulk selection experiments (table S2).
In vitro P. falciparum cell culture P. falciparum ABS parasites were cultured in human O + RBCs obtained from the Interstate Blood Bank (Memphis, TN) at 3% hematocrit, using RPMI 1640 medium supplemented with 25 mM Hepes, sodium bicarbonate (2.1 g/liter), gentamicin (10 μg/ml), 50 μM hypoxanthine, 0.5% (w/v) AlbuMAX II (Thermo Fisher Scientific), and 10% (v/v) human serum (Interstate Blood Bank).Parasites were maintained at 37°C under 5% O 2 /5% CO 2 /90% N 2 gas conditions.Single clones were obtained from the bulk progeny pools, the RF7 parent, and the HapU progeny by limiting dilution cloning at 0.5 parasites per well.Culture medium was replenished twice a week, and fresh RBCs were added once per week.Clones were expanded from days 19 to 35 after parasitemias were above the detection threshold, and the pm2 copy number was measured by TaqMan-based real-time polymerase chain reaction (PCR) on isolated genomic DNA (gDNA; see the section below).Protocol AAAU3761, describing the use of human RBCs from anonymized, de-identified donors for P. falciparum cultures, was approved by the Columbia University Institutional Review Board, who deemed this to be not human subjects research under 45 CFR 46.

Drug treatments applied on bulk progeny pools
We applied 75 nM PPQ (5.6× IC 50 for NF54), 200 nM mdCQ (9.1× IC 50 for NF54), and 1 μM PYM (54× IC 50 for NF54) for 72 to 96 hours on bulk progeny pools in 10-ml culture volumes at 3% hematocrit and ~1% parasitemia.Drug was replenished every 24 hours with media changes.Following 72 or 96 hours of drug exposure, drug removal was achieved through centrifugation of the culture at 1500 rpm for 5 min, discarding the supernatant and resuspending the cultures in 50 ml of fresh media, with three washes in total.Cultures were monitored daily and harvested for gDNA once parasitemia reached at least 1%, which was 8 days from the start of treatment for PYM and mdCQ and 16 days for PPQ.On the basis of our initial tests, we found that these drug exposure conditions effectively cleared NF54 but not RF7 parasites and thus proved suitable to eliminate selfed NF54 progeny, thereby increasing the numbers of independent recombinant progeny.We did not subject these bulk progeny pools to DHA pressure because the other drug selections already yielded progeny with diverse DHA phenotypes, as measured using RSA and AUC values.

WGS of bulk pools and individual progeny
Parasite cultures at 1 to 5% parasitemia were lysed with 0.1% saponin in 1× phosphate-buffered saline (PBS), and gDNA was then extracted using the QIAamp DNA Blood Mini Kit (Qiagen).DNA concentrations were determined using the Qubit dsDNA BR assay kit (Thermo Fisher Scientific).WGS libraries were prepared using the Illumina Nextera DNA Flex protocol.Briefly, 150 ng of gDNA was fragmented and tagmented using bead-linked transposons and subsequently amplified by 5 cycles of PCR to add dual-index adapter sequences to the DNA fragments.The libraries were quantified, pooled, and sequenced on Illumina MiSeq or NextSeq platforms to generate 300-or 150-bp pairedend reads, respectively.Drug-selected bulk cultures yielded a mean 44× depth of coverage by Illumina MiSeq sequencing.Cloned progeny were subjected to WGS by Illumina NextSeq or MiSeq to identify selfed versus genetic recombinants.Thereafter, representative progeny from each of the 34 recombinant haplotypes, as well as the two parents RF7 and NF54, were resequenced on the Illumina MiSeq platform to obtain between 2.7 and 4.6 M reads, corresponding to an average of 40.7× depth of coverage, with on average at least 94.1% of each genome having 10 or more reads (table S3).Sequence data were aligned to the P. falciparum 3D7/ NF54 genome (PlasmoDB version 36.0) using Burrow-Wheeler Alignment.Reads that did not map to the reference genome, as well as optical duplicates, were removed using SAMtools and Picard.The reads were realigned around indels using Genome Analyses Tool Kit (GATK) RealignerTargetCreator, and base quality scores were recalibrated using GATK BaseRecalibrator.
All variants were called using mpileup and were filtered on the basis of quality scores (minimum base quality score ≥ 20, mapping quality > 30, read depth ≥ 5) and multiallelic = FALSE to obtain high-quality SNPs, which were then annotated with snpEFF.SNPs for highly polymorphic surface antigens and multigene families (notably var, stevor, and rifin, located mostly at the subtelomeric ends of chromosomes) were excluded as these are prone to stochastic changes during in vitro culture.To obtain the list of SNPs that differed between the parents in the core genome, we retained only homozygous SNPs, based on >90% alternate allele frequency in RF7 and >90% reference allele frequency in NF54.For QTL mapping, pm2/3 copy was manually imputed as an additional marker "Pf3D7_14_v3_295261_PM2."This analysis gave a total of 18,490 SNPs (18k SNP set).SAMTools mpileup was used to find SNPs in the 18k SNP set for progeny or in targeted genomic regions including genes mediating known drug resistance phenotypes.SNPs that showed an allele frequency between 10 and 90% in the progeny clones were classified as heterozygous and defined as "missing/undetermined."Bayesian information criterion (BIC)-Seq was used to check for copy number variations using the Bayesian statistical model (78).Copy number differences between RF7 and NF54 were only observed in a chr14 segment harboring the tandem pm2/3 genes.

Microsatellites for identifying recombinant progeny and for determining PfPailin genotypes within the k13flanking regions
To provide a microsatellite (MS) signature for each recombinant progeny, we established a list of MS markers that differentiated RF7 from NF54.For this analysis, we wrote a custom Python script, using the pysam module, to call WGS reads showing insertions or deletions at targeted genomic loci (79).Integrated Genome Viewer was used to visually verify the presence of MS markers.The markers used herein were C2M18, TAA81, BM5, C3M67, TA1, and C13M87 (fig.S14).To identify MS markers surrounding the k13 gene, we referred to markers previously profiled in PfPailin parasites that spanned −50.0 to +31.5 kb of this gene (5,80) and determined these marker sizes in the progeny based on WGS reads (fig.S3).

Confirmation of progeny clones via MS-based multiplexed fragment analysis
We used multiplexed fragment analysis to validate the genetic purity of individual progeny clones, pre-and postphenotyping (fig.S14).PCRs were set up individually for C2M18, TAA81, BM5, C3M67, TA1, and C13M87, using forward primers fluorescently labeled with 6-FAM (blue), ATTO550/NED (yellow), or ATTO532/VIC (green) as described previously (79).PCR products treated with ExoSAP-IT Express Reagent (Thermo Fisher Scientific) were sent for capillary electrophoresis with the Liz500 ladder as reference (Genewiz).Data were analyzed using Peak Scanner 2 (Thermo Fisher Scientific) to determine MS sizes, with RF7 or NF54 included as references.

Phenotypic assessment of drug sensitivities in genetic cross parents and progeny clones Drug susceptibility assays for RF7 × NF54 parents
To profile the genetic cross parents' susceptibility to a panel of 30 antimalarial drugs or preclinical compounds, we performed 72hour dose-response assays on sorbitol-synchronized ring-stage cultures in 3 to 17 independent experiments with technical replicates.Assays (with 0.4 to 0.7% starting parasitemia and 1% hematocrit) used a 10-point twofold serial dilution of the compounds and included 0.1% dimethyl sulfoxide (DMSO) control-treated samples.

DHA and PSAs
To obtain tightly synchronized parasites for RSAs and PSAs, cultures were treated with 5% D-Sorbitol (Sigma-Aldrich) for 15 min at 37°C to remove mature parasites.For RSAs, we then cultured for 40 hours and purified late-stage segmented schizonts over 35 and 65% Percoll double density gradients (81).Purified schizonts were incubated with fresh RBCs for 3 hours and treated with 5% D-Sorbitol to obtain 0-to 3-hour postinvasion early rings.These rings were exposed to DHA concentrations ranging from 5.5 to 2,800 nM, prepared as twofold serial dilutions in 96-well plates.Wells were inoculated with 200-μl samples (at 1% starting parasitemia and 1% hematocrit), with technical duplicates.Control wells contained 0.1% DMSO.After a 4-hour incubation at 37°C, wells were washed three times with complete medium to remove drug, and contents were transferred to fresh 96-well plates using the Freedom EVO MCA96 liquid-handling instrument (Tecan).Cultures were subsequently maintained for an additional 66 hours in drug-free medium.For PSAs, sorbitol-synchronized ring-stage parasites were exposed for 72 hours to 200 nM PPQ (dissolved in 0.5% lactic acid), alongside drug-free wells, in technical duplicates.

PPQ, LMF, MFQ, MMV665939, and MMV675939 drug susceptibility assays
For the PPQ AUC measurements, we performed 72-hour assays with synchronized ring-stage parasites exposed to PPQ at twofold serially diluted concentrations ranging from 0.8 nM to 25.6 μM, alongside drug-free wells, in technical duplicates.Standardized 72-hour dose-response assays were conducted similarly for LMF, MFQ, MMV665939, and MMV675939 at 10-point concentrations starting from 600 nM, 200 nM, 32 μM, and 16 μM for these compounds, respectively.DMSO-treated control wells (0.1%) were run in parallel.For all drug assays, the RF7 and NF54 parents were included as reference controls.

Flow cytometry for quantification of viable parasites
Parasitemia for drug-treated and vehicle control-treated wells were measured at 72 hours by flow cytometry, as previously described (15).Briefly, parasites were incubated with 1× SYBR Green (Thermo Fisher Scientific) and 100 nM MitoTracker DeepRed (Thermo Fisher Scientific) for 30 min at 37°C, followed by quenching with 1× PBS.On average, we analyzed 10,000 cells per sample, using an iQue Screener Plus cytometer (Sartorius).Viable parasites were defined as the percentage of MitoTracker-positive-and SYBR Green-positive-labeled cells.

Calculation of RSA, PSA, AUC, and IC 50 values
For all drug assays conducted here, we included kill controls in which 1 μM DHA-treated parasites were used as a background control to achieve complete parasite killing and subtracted this percent parasitemia from the parasitemias measured for each well.Parasite survival in the presence of DHA or PPQ was expressed as the percentage of the background-subtracted parasitemia in the 700 nM DHA-treated samples or 200 nM PPQ-treated samples divided by the parasitemia of the DMSO-treated samples.Mean RSA survival rates of >1% were defined as DHA resistant (82).The AUC values for DHA, PPQ, and MMV675939 were calculated on the basis of total parasite survival across the range of 21.9 nM to 2.8 μM for DHA, 3.1 nM to 25.6 μM for PPQ, and 31.3 nM to 16 μM for MMV675939.IC 50 values were determined by applying a nonlinear regression model (sigmoidal dose-response with variable slope) on the normalized % survival across the log-transformed drug concentrations, using Prism v8.3.1 (GraphPad).

Genetic map construction
Duplicate markers were removed to obtain 2,091 markers using R/ qtl (83).These markers were analyzed in JoinMap v5 (84) using the independence LOD parameter to generate a single linkage group for each chromosome, except for chr12 where there were three linkage groups.Using the resulting 1918 markers, we constructed a genetic map for RF7 × NF54 using Kosambi's regression mapping function and adjusted for linkages with a recombination frequency < 0.5 and LOD scores > 0 (table S6).

Bulk segregant analysis of bulk pools and drugselected clones
Identification of QTLs in bulk pools treated with each of the drug conditions was conducted on the 18k SNP set obtained from WGS data using G 0 and Quantitative trait locus (QTL)-seq approaches in the QTLseqr package in R (85).A sliding window of 100 kb was used for calculating the tricube-smoothed G 0 and ∆(SNP index) values of each SNP within that window.False discovery rate q values were adjusted and customized for each drug-to-drug or drug-to-control pairwise analysis, which enabled us to account for the differences in sequencing coverage and the varying degrees of loci enrichment between drug-selected bulk pools.Outlier SNPs were filtered by Hampel's rule (85).In addition to the bulk segregant analysis conducted between drug-treated bulk pools (Fig. 2) and between drug-treated and untreated bulk pools (fig.S2), we performed a similar analysis on progeny clones from bulk progeny pools that were exposed to drug or derived in the absence of drug.For this analysis, progeny clones selected by PPQ, mdCQ, PYM, or untreated controls (N = 12 for untreated, N = 12 for PYM, N = 7 for mdCQ, and N = 6 for PPQ) were grouped, and the RF7 allele frequency within each group of clones was averaged, thereby generating pooled clones that was used for bulk segregant analysis (Fig. 3E).

Individual clone-based quantitative trait loci mapping
After filtering out SNPs that were missing in >5 of the 49 progeny (representing the 34 haplotypes) profiled for drug susceptibilities, we retained 15,869 SNPs and used these with the R/qtl2 package (86) to map QTL peaks.To identify significant QTLs for each phenotypic output, permutations of phenotypic data were performed 1000 times to obtain a distribution of maximum LOD scores.These scores were then used to calculate the LOD threshold at 95% probability.Fine-mapping of the QTL segments was subsequently performed using Bayesian interval mapping at a 95% confidence level.To elucidate secondary QTLs, LOD scores were calculated after setting the major peak as an additive covariate.The LOD threshold at 95% probability was then recalculated.The k13, pfcrt, and pm2/3 genes were applied separately as additive covariates for each of the QTL analyses.The percent variance associated with each QTL was determined by establishing linear models for each QTL and comparing additive versus interactive QTL effects in R/qtl (83).

Gene editing of pfcrt and k13 by ZFN-based and/or CRISPR-Cas9 approaches
We performed CRISPR-Cas9 editing to remove the C580Y mutation from the k13 locus in RF7 clones harboring either one or three copies of pm2/3, using the all-in-one plasmid, pDC2-cam-coSpCas9-U6-gRNA-k13_bsm-hdhfr (12).We also used zincfinger nuclease-mediated editing to replace the 3D7 WT pfcrt allele with the Dd2 + M343L pfcrt isoform (carrying nine point mutations) in HapD progeny, as previously described (87).CRISPR-Cas9 editing of pfcrt to remove the M343L single-point mutation in the HapU progeny harboring either 1, 2, or 3 copies of pm2/3 was performed using the all-in-one plasmid, pDC2-cam-Cas9-U6-gRNA-pfcrt_M343-bsd.Plasmid cloning was performed as for k13, except that the guide RNA (gRNA) was inserted using In-Fusion cloning (Takara), and blasticidin S-deaminase (bsd) was used as the drug selection cassette instead of human dhfr.gRNAs were cloned using the primer pair p8234 + p8235.Donor templates were amplified and cloned into the final vector using the primer pairs p8236 + p8237.Site-directed mutagenesis was performed using the allele-specific primer pairs p8238 + p8239.Donor inserts were PCR-amplified using the primer pair p7949 + p7950 and sequenced with the internal primer p7989.The primer sequences are listed in table S13.
To generate gene-edited lines, ring-stage parasites at 5 to 10% parasitemia were electroporated with 50 μg of circular plasmid DNA resuspended in Cytomix.Transfected parasites were selected by culturing in the presence of WR99210 (Jacobus Pharmaceuticals) or blasticidin for 6 to 8 days post-electroporation.The RF7 clones were subjected to 5 nM WR99210, while the HapD and HapU parasites were selected under 2.5 nM WR99210 or blasticidin (2 μg/ml).Parasite cultures were monitored for recrudescence by microscopy for up 8 weeks post-electroporation.To screen for successful editing, the k13 and pfcrt loci were amplified directly from parasite cultures using the MyTaq Blood-PCR Kit (Bioline) (79).PCR products were submitted for Sanger sequencing, and edited parasites were cloned by limiting dilution.Gene-edited lines and their phenotypes are described in table S8.
Quantitative PCR for determination of pm2 copy number Multiplexed TaqMan quantitative PCR (qPCR) of pm2 labeled with FAM and single-copy β-tubulin labeled with HEX (as an internal control) were performed on gDNAs extracted from the two parents, progeny clones, and gene-edited lines, pre-and postphenotyping.Reactions were run on a QuantStudio 3 real-time PCR system (Thermo Fisher Scientific) (79).Five standards of gene fragments, mixed at 1:1, 2:1, 3:1, 4:1, and 5:1 molar ratios of pm2:βtubulin, as well as the 3D7 line (one copy of pm2), were included as copy number controls.Each sample was assayed in technical duplicates.The pm2 copy number for each progeny line was calculated by normalizing to the 3D7 control, using the relative quantification method (79).Primer sequences are listed in table S13.

Competitive fitness assays for k13 and pm2/3 Isogenic cocultures
Fitness assays were performed by coculturing isogenic parasite lines in 1:1 ratios.These assays paired isogenic k13 WT versus mutant clones in RF7 clones that harbored either one or three copies of pm2/3.We also paired single versus multicopy pm2/3 RF7 clones.These three pairwise cocultures were used to independently examine k13 genotype or pm2/3 copy number.Assays were initiated with tightly synchronized rings and conducted on four independent occasions with technical replicates.Cultures were maintained at 6ml volumes over a period of 40 days (~20 generations), and every 4 days a fraction of each coculture was harvested for gDNA as described above.

k13 allelic discrimination qPCR assays
The percentage of WT or mutant k13 alleles in each sample was determined using custom TaqMan fluorescence-labeled minor groove binder (MGB) probes in TaqMan allelic discrimination real-time PCR assays, as described previously (12).Probes were designed to specifically detect either the k13 C580Y propeller mutation (HEX probe, Eurofins) or the WT allele (FAM probe, Eurofins).Primer sequences are listed in table S13.Mixtures of WT and mutant plasmids in fixed ratios (0:100, 20:80, 40:60, 50:50, 60:40, 80:20, and 100:0) were run as parallel controls to ensure accurate allelic detection.qPCR reactions for each sample were run in duplicate, with each 20-μl reaction consisting of 1× QuantiFAST reaction mix containing ROX reference dye (Qiagen), 0.66 μM forward and reverse primers, 0.16 μM FAM-MGB and HEX-MGB TaqMan probes, and 10 ng of gDNA.Cycling conditions were as follows: 1 cycle of 30 s at 60°C and 5 min at 95°C and 40 cycles of 30 s at 95°C and 1 min at 60°C.Every assay included no-template negative controls as well as positive controls (mixtures of WT and mutant plasmids in fixed ratios), which were run in triplicate.Amplification and detection of fluorescence were carried out on a QuantStudio3 real-time PCR system using the genotyping assay mode.Rn (the fluorescence of the FAM or HEX probe) was normalized to the fluorescence signal of the ROX reporter dye.Background-normalized fluorescence (Rn minus baseline or ΔRn) was calculated as a function of cycle number.pm2/3 allelic discrimination qPCR assays TaqMan allelic discrimination was not suitable to quantify the percentage of single versus two or more copies of pm2/3 as the multicopy alleles have identical sequences.To facilitate the differentiation of RF7 clones expressing one or three copies of pm2/3, we therefore conducted WGS of pm2/3 multicopy versus single-copy RF7 clones.This analysis identified a SNP (gaA/gaT) present within the isoleucine-tRNA ligase gene (Pf3D7_1332900), corresponding to an E459D mutation in the RF7 clone harboring a three-copy pm2/3 amplicon, presumably as a result of genetic drift during extended culture and clonal expansion.Forward and reverse PCR primers and custom TaqMan fluorescence-labeled probes were designed to target the E459 WT (FAM, Eurofins) or E459D mutant (HEX, Eurofins) alleles for this gene (table S13).
The efficiency and sensitivity of these TaqMan primers were assessed using standard curves comprising 10-fold serially diluted templates ranging from 10 to 0.001 ng.Robustness was demonstrated by high efficiency (94 to 96%) and R 2 values (0.99 to 1.00).The quantitative accuracy in genotype calling was assessed by performing multiplex qPCR assays using mixtures of WT and mutant plasmids in predefined fixed ratios (0:100, 20:80, 40:60, 50:50, 60:40, 80:20, and 100:0).These mixtures were obtained by cloning a ~800-bp fragment for isoleucine-tRNA ligase centered around the mutant or WT SNP in pGEM plasmids.Triplicate data points clustered tightly and gave expected ratios of WT to mutant alleles, indicating high reproducibility in the data across the fitted curve (R 2 = 0.86 to 0.87).Cocultures consisting of the RF7 isogenic clones containing multicopy pm2/3 with the isoleucine-tRNA ligase E459D mutation, and the single-copy pm2/3 and WT E459 allele, were initiated at a 1:1 ratio, and fitness assays were conducted over a 40-day period.qPCR reactions were run with probes targeting the isoleucine tRNA ligase marker that was diagnostic for single versus multicopy pm2/3, using conditions described above.
To determine the WT or mutant allele frequency in each sample, we retained only ΔRn values in samples where the threshold cycle (C t ) was at least 3 cycles less than the no-template control.Next, we subtracted the ΔRn of samples from the ΔRn of the no-template negative control.We subsequently normalized the fluorescence to 100% using the positive control plasmids (tested in parallel reactions) to obtain the percentage of the WT and mutant alleles for each sample.The final percentage of the mutant allele was defined as the average of two values: the normalized percentage of the mutant allele and 100% minus the normalized percentage of the WT allele.

Fitness cost assessment
The fitness cost associated with the k13 mutation or multicopy pm2/ 3 was calculated relative to its isogenic WT or single-copy counterpart, respectively, using the following equation: P 0 = P × (1 − f ) n .P 0 is the percentage of line expressing the k13 C580Y mutation or multicopy pm2/3 at the assay endpoint.P is the percentage of the coculture that expresses the k13 WT allele or single-copy pm2/3 on day 0. n is the number of generations from the assay start to finish, and f is the fitness cost.For the TaqMan allelic discrimination real-time PCR assays described above, n was set to 20.

CQ or PPQ uptake assays in PfCRT-containing proteoliposomes
Inhibition of CQ or PPQ uptake was measured using proteoliposomes containing PfCRT, as described previously (32) with the following modifications.Purified PfCRT variants were reconstituted in preformed liposomes made of Escherichia coli total lipids:cholesteryl hemisuccinate in a ratio of 97:3 (w/w), and the lumen of the proteoliposomes was composed of 100 mM KP i , (pH 7.5) and 2 mM β-mercaptoethanol.Uptake of 100 nM [ 3 H]-CQ (1 Ci/mmol) or [ 3 H]-PPQ (1 Ci/mmol) was performed by diluting PfCRT-containing proteoliposomes (30 ng of PfCRT per assay) in 50 μl of 100 mM tris/MES (pH 5.5) in the absence or presence of 1 μM atovaquone, CQ, PPQ, MMV665939, MMV675939, or verapamil.Valinomycin (1 μM) was added to the reaction to generate a K + diffusion potential-driven membrane potential.After 30 s, the reactions were quenched by the addition of ice-cold 100 mM KP i (pH 6.0) and 100 mM LiCl before filtration.The radioactivity retained on the filters was determined with scintillation counting in a Hidex SL300 scintillation counter.Proteoliposomes that contain the CQresistant 7G8 or the PPQ-resistant 7G8 + F145I or 7G8 + C350R PfCRT isoforms were used herein.No drug ("−") and atovaquone were included as negative controls, for which we expected no or minimal inhibition of CQ or PPQ uptake.CQ or PPQ was used as the positive control as PPQ has been reported to interact with PfCRT and inhibit binding of CQ (32).Data were normalized to the specific signal (total counts per minute in PfCRT-containing proteoliposomes minus counts per minute in liposomes devoid of PfCRT).Means ± SE values are presented as the percentage of the uptake obtained in the absence of drug ("−").Data were collected from three independent experiments with technical duplicates.

Statistical analysis
Unless stated otherwise, experiments were performed on at least four independent occasions, and exact sample sizes are given in the figure legends or supplementary tables and figures.Scatterplots and bar graphs are presented as means ± SE.Data were analyzed and plotted using Prism v8.3.1 (GraphPad).QTL mapping was analyzed and plotted in R using R/qtl, R/qtl2, and ggplot2.Comparisons between two groups were analyzed by Mann-Whitney U tests if at least four repeats were obtained per group or unpaired Student's t test if otherwise.Statistical significance was defined by two-sided P values (*P < 0.05, **P < 0.01, and ***P < 0.001).P values less than 0.05 were considered statistically significant.ns indicates no significant difference.

Supplementary Materials
This PDF file includes: Figs.S1 to S14 Legends for tables S1 to S13 Other Supplementary Material for this manuscript includes the following: Tables S1 to S13

Fig. 1 .
Fig. 1.Antimalarial susceptibilities of the genetic cross parents RF7 × NF54 and the experimental workflow for bulk selection versus individual clone-based linkage mapping.(A) Bar plots of % survival derived from the DHA ring-stage survival assay (RSA) or the PPQ survival assay (PSA) and the area under the curve (AUC) values, as determined for the genetic cross parents RF7 and NF54.In the RSAs and PSAs, early ring-stage parasites were exposed to the pharmacologically relevant concentrations of 700 nM DHA for 4 hours or 200 nM PPQ for 72 hours, respectively, and survival was measured as a percentage of mock-treated cultures.(B) IC 50 values of RF7 and NF54 parasites exposed for 72 hours to first-line antimalarials.Values represent the means ± SE (N = 3 to 17, n = 2).Statistical significance was determined by unpaired Student's t tests, with a Holm-Sidak post hoc test to correct for multiple comparisons.The IC 50 fold shifts are indicated above the bars.*P < 0.05, **P < 0.01, ***P < 0.001.Numbers listed above the statistics indicate the fold differences between parental IC 50 values.The table summarizes parental phenotypic and genotypic characteristics.WT, wild-type.Mutant genotypes include the following: Dd2 pfcrt mutations: M74I, N75E, K76T, A220S, Q271E, N326S, I356T, and R371I; quadruple dhfr mutations: N51I, C59R, S108N, and I164L.(C) Overview of the genetic cross pipeline performed in humanized FRG-NOD mice.huRBCs, human red blood cells.(D) Bulk selection approach used for bulk segregant analysis.(E) Individual clone-based linkage approach used to identify genetic determinants of drug resistance.Images were created with BioRender.com.

Fig. 3 .
Fig. 3. Genetic analysis of recombinant progeny and enrichment of diverse progeny using drug pulses.(A) Number and percentage of selfed progeny versus genetic recombinants obtained from cloning in the absence of drug or following drug exposure.Distribution of the 34 unique recombinants obtained with no drug and/or selected by PPQ, mdCQ, or PYM.(B) Allelic map for the 34 recombinant progeny and two parents, grouped by drug treatment condition.Shown are the parental alleles inherited by each progeny for 14,476 SNPs (after excluding SNPs missing in any of the 34 recombinant haplotypes), the genotypes of known resistance markers, and the number of derived progeny clones per haplotype sorted by selection condition.(C) Mean RF7 allele frequency inherited by the 34 recombinant progeny clones for each drug treatment group.(D) Count of KEL1/PLA1/PfPailin haplotypes in progeny clones that were selected by each drug condition.Statistical significance was tested for the drug-selected progeny clones against the "no drug" group, using Fisher's exact test.**P < 0.01; ***P < 0.001; ns, not significant.(E) Frequency of RF7 parental allele in progeny clones derived posttreatment with PPQ (N = 6), mdCQ (N = 7), or PYM (N = 12) or in the absence of drug (N = 12).Individual allele frequencies (light dots in the background) are reported as a weighted allele frequency across the number of clones for each group.The solid colored lines represent the averaged RF7 allele frequency across windows of 100 consecutive SNPs.The skews in allele frequency at specific loci in the genome are indicated by dashed lines.

Fig. 4 .
Fig. 4. Phenotypic response of parents and progeny to DHA and PPQ.(A and B) Dose-response curves for RF7 and NF54 parents across a range of DHA and PPQ concentrations (N = 3, n = 2).The percent survival at the RSA and PSA concentrations of 700 nM DHA and 200 nM PPQ for 4 or 72 hours, respectively, were used to determine DHA and PPQ resistance levels, whereas AUC values were measured as total survival across a range of concentrations (22 nM to 2.8 μM for DHA and 1.6 nM to 25.6 μM for PPQ).(C and D) DHA and PPQ response measured by %RSA and DHA AUC values (C) and %PSA and PPQ AUC values (D), respectively, in the 34 independent recombinant haplotypes and two parents.Progeny clones were obtained either in the absence of drug pressure (gray) or after selection with PYM (purple), mdCQ (orange), or PPQ (green).Data are plotted alongside RF7 (red arrow) and NF54 (blue arrow).Each bar represents the mean percent survival ± SE or AUC ± SE for a recombinant haplotype profiled in four independent experiments with technical duplicates.Haplotypes are ordered by the parasites' resistance levels for each drug metric.For certain haplotype groups in which two or more identical clones were obtained, multiple points depict phenotypic results for more than one sibling progeny.

Fig. 5 .
Fig. 5. QTL mapping of DHA responses in progeny identifies k13 as the primary DHA resistance locus.(A and B) LOD plots for %RSA and AUC levels showing the significant QTLs above the 95% probability threshold (red line).(C) List of QTL segments for %RSA and AUC levels.(D) Genes in QTL segment on chr13, with those having nonsynonymous mutations between RF7 and NF54 colored in orange (n = 20) or gray where mutations were absent in RF7 and NF54.Gene names are as listed.uf, unknown function.(E) The %RSA and AUC levels in recombinant progeny segregated by k13 parental allele, C580Y (in RF7) and WT (in NF54).Significance was tested using Mann-Whitney U tests.***P < 0.001.(F) DHA responses in k13-edited isogenic RF7 clones showed that the k13 WT allele reduces %RSA and AUC levels.Bars represent the means ± SE (N = 4, n = 2).Significance was tested using Mann-Whitney U tests.*P < 0.05.(G and H) Linkage of chr1 (G) and chr14 (H) segments with pfcrt from QTL analysis using the pfcrt genotype as an outcome.Shown are the significant QTLs above the 95% probability threshold for each analysis.(I) LOD plot for AUC levels after adjusting for k13 as a covariate suggests independent inheritance of the chr14 segment with k13 and coinheritance of the chr1 segment with k13.(J) Scatterplot of AUC levels in progeny segregated by k13, mrp1 (chr1), and arps10 (chr14) genotypes, showing that the RF7 chr14 segment is associated with increased AUC levels.We note that none of the progeny harbored MUT mrp1 and WT k13 alleles.Significance between groups was tested using Mann-Whitney U tests.*P < 0.05, ***P < 0.001.MUT, mutant.

Fig. 7 .
Fig. 7. Impact of mutant k13 and multicopy pm2/3 on parasite asexual fitness.(A) Experimental design showing the generation of the panel of isogenic RF7 lines used in the coculture competitive fitness assays.(B) Pairwise competitive growth assays showing the proportion of k13 C580Y or WT alleles in the RF7 line harboring either three copies or one copy of pm2/3 and the proportion of RF7 lines carrying either three copies or one copy of pm2/3.Values shown are the averaged percentages from four independent experiments with two technical replicates per pairwise competition assay.(C) Fitness cost per generation, showing the relative change in percentage of the k13 and pm2/3 genotypes for each pairwise comparison.

Fig. 8 .
Fig. 8. QTL mapping of LMF and MFQ response identifies multiple peaks including mutant Dd2 + M343L pfcrt.(A) LMF IC 50 and (B) MFQ IC 50 values of the 34 unique recombinant haplotypes and two parents.Each bar represents the mean ± SE IC 50 for a recombinant haplotype (N = 4, n = 2).Haplotypes were ordered by LMF IC 50 levels.For certain haplotype groups in which identical clones were obtained, multiple points depict data for more than one sibling progeny.Insets depict LMF and MFQ dose-response curves for RF7 (red) or NF54 (blue).Each curve represents the mean ± SD for an independent experiment.(C and D) LOD plots showing QTLs for LMF and MFQ above or near the 95% probability threshold (red line).(E and F) LMF and MFQ IC 50 values for the 34 recombinant progeny and two parents, grouped by their pfcrt haplotypes (E) or grouped by their atg18 and pfcrt haplotypes (F).MUT, mutant T38I.(G) LMF IC 50 and (H) MFQ IC 50 values in pfcrt-edited HapD progeny (3D7 versus Dd2 + M343L) having single pm2/3 copy, and in edited HapU progeny (Dd2 versus Dd2 + M343L) having either one, two, or three copies of pm2/3.Bars represent the means ± SE (N = 3 to 4, n = 2).The colored key for parasite lines applies to (G) to (K).(I) LMF and (J) MFQ dose-response curves in edited HapD progeny showing that mutant Dd2 + M343L pfcrt confers increased tolerance to these drugs.Each line depicts the mean ± SE percent parasite survival (N = 3 to 4, n = 2).(K) MFQ dose-response curve in pfcrt-edited HapU progeny with variable pm2/3 copies, showing that single pm2/3 copy associates with reduced MFQ sensitivity.Each line depicts the mean ± SE percent parasite survival (N = 4, n = 2).Statistical significance was determined by Mann-Whitney U tests (A, B, E, and F) or unpaired Student's t tests (G and H).*P < 0.05, **P < 0.01, ***P < 0.001.