Replacement of the mutant pfcrt T76 codon with the wild-type K76 codon, in CQ-resistant parasites harboring pfcrt alleles representative of Old World (Dd2) or New World (7G8) origins, was achieved by allelic exchange. The strategy was based on homologous recombination with the transfection plasmid pK76, effectively replacing endogenous pfcrt with a recombinant locus consisting of a downstream, full-length functional pfcrt and an upstream, truncated nonfunctional pfcrt remnant (Figure 1A). Crossover events downstream of codon 76 introduced a T76K substitution in the full-length functional pfcrt, generating a ‘back-mutant' line (Table I and Figure 1A). Crossover events upstream of codon 76 generated the same recombinant locus without altering the haplotype, thus providing recombinant controls.

Minimal promoter elements were first defined that could drive transcription of recombinant full-length pfcrt. For this, we amplified the 3.0 kb region separating pfcrt from the upstream cg3 gene. Unidirectional digestion produced sequences stretching 1.6, 1.2 and 1.0 kb upstream from the pfcrt start codon. These elements gave mean±s.e.m. luciferase activities of 84±12, 60±13 and 3±2%, respectively, relative to the 3.0 kb element (n=4 assays). Our allelic exchange strategy employed the 1.6 kb element, which gave acceptable transcription activity while retaining favorable odds that a recombination event would occur downstream of codon 76.

To generate the desired recombinant clones, we transfected Dd2 and 7G8 parasites with pK76 (Figure 1A). Parasite lines that underwent integration into pfcrt were identified by PCR and sequencing, and cloned. Recombinant back-mutants were named T76K-1^Dd2, T76K-2^Dd2, T76K-1^7G8 and T76K-2^7G8, based on their genetic background (Table I). Recombinant controls were named C-1^Dd2, C-2^Dd2, C-1^7G8 and C-2^7G8. PCR with primers p3/p4 and p5/p6 yielded, for all recombinant lines, 3.4 and 3.3 kb bands corresponding to functional and truncated pfcrt sequences, respectively (Figure 1B). Sequencing of nested PCR products confirmed their PfCRT haplotypes. Primers p5/p4, specific for endogenous pfcrt, produced a 3.4 kb band from parental (nontransfected) lines and episomally transfected cultures but not from the cloned recombinants (Figure 1B; data not shown).

Allelic replacement was confirmed by Southern hybridization. Upon BglII/StuI/XbaI digestion, all recombinant lines displayed 7.8, 7.4 and 4.8 kb bands when hybridized with a pfcrt probe (exon 2–intron 3), and 7.8 and 7.4 kb bands with a blasticidin S-deaminase (bsd) probe (Figure 1C). Parental lines produced a 5.2 kb band. StuI/BglII digestion, which cleaves sites flanking pfcrt (Figure 1A), revealed tandem integration of two or more plasmid copies in all recombinant lines (Figure 1D).

RT–PCR with primers p7/p9 showed transcription from the full-length pfcrt locus in all lines (Figure 1E). In the recombinants, transcription was also detected with primers p7/p6 from the upstream remnant, which retained the endogenous promoter but was truncated midway through pfcrt (Figure 1E). Northern blot hybridizations revealed a 3.5 kb transcript emanating from the functional locus in all recombinant lines, compared to a 4.2 kb transcript from parental Dd2 and 7G8 (Figure 1F). Based on earlier mapping (Waller et al, 2003), this indicated that recombinant downstream pfcrt transcripts originated close to the start of the 1.6 kb element. Hybridizations with ef-1α revealed similar loading in all lanes. Densitometry predicted a 65–75% decrease in steady-state transcript levels of recombinant functional pfcrt compared to the parental locus. Western blots of highly synchronized parasite protein extracts, probed with antibodies to either PfCRT or PfERD2 (loading control), predicted a 40–55% decrease in expression levels in back-mutant and control recombinants, compared to parental lines (Figure 1G). This greater reduction in mRNA compared to protein levels upon pfcrt allelic exchange was also observed previously (Waller et al, 2003).

Generating the back-mutant lines enabled us to determine the contribution of K76T to P. falciparum susceptibility to antimalarial agents. Assays were performed against CQ and its in vivo metabolite mono-desethylchloroquine (m-dCQ), quinine (QN) and its diastereoisomer quinidine (QD), amodiaquine (ADQ), mefloquine (MFQ), artemisinin (ART) and amantadine (AMT). CQ, m-dCQ and QN were all tested with or without VP (Figures 2 and 3). For each line, 3–14 independent assays were performed in duplicate (detailed after in Supplementary Table I).

For the Dd2 lineage, the T76K substitution resulted in a total loss of CQR. Mean CQ IC₅₀ values were 22.1 and 25.7 nM in the back-mutants T76K-1^Dd2 and T76K-2^Dd2, a degree of sensitivity equivalent to the reference CQ-sensitive line GC03 (22.8 nM; Supplementary Table I). These values were six- to seven-fold lower than those of C-1^Dd2 and C-2^Dd2 controls (P<0.01) and 11- to 13-fold lower than Dd2. Loss of resistance was even more dramatic with m-dCQ, with T76K-1^Dd2 and T76K-2^Dd2 demonstrating IC₅₀ values of 23.8 and 30.8 nM, directly comparable to GC03 (35.3 nM). These back-mutant m-dCQ values were 24- to 38-fold lower than recombinant controls (P<0.001) and 48- to 62-fold lower than Dd2. We note that the control lines C-1^Dd2 and C-2^Dd2 (which underexpress PfCRT by about half; Figure 1G) had CQ and m-dCQ IC₅₀ values that were approximately two-fold lower than Dd2. This confirmed an earlier observation that reduced PfCRT expression levels can lower the degree of CQR (Waller et al, 2003).

Interestingly, there was a statistically significant, approximately 40% decrease in IC₅₀ values for QN in the Dd2 recombinant control and back-mutant lines (which all gave similar IC₅₀ values), compared to Dd2 (Figure 2). Smaller decreases in IC₅₀ values were observed for QD, MFQ and ADQ in the Dd2 back-mutants; however, these were not significant.

We also assayed AMT, in view of recent reports that CQ-resistant parasites are hypersensitive to this influenza A M2 ion channel inhibitor (Johnson et al, 2004). Dd2 and recombinant control lines showed mean IC₅₀ values of 20–29 μM (Figure 2). In comparison, the CQ-sensitive Dd2 back-mutants became 11- to 17-fold less susceptible (P<0.001; mean IC₅₀ values of 309–331 μM). Thus, in Dd2 parasites, the PfCRT K76T mutation appears to significantly influence AMT susceptibility. We note that in a previous study, in vitro AMT pressure of the K1 (Thailand) CQ-resistant line resulted in an AMT-resistant mutant that had also lost VP-reversible CQR and had acquired two novel pfcrt point mutations (T152A and S163R) (Johnson et al, 2004). These data suggest that various mutations in Dd2-like pfcrt alleles can confer parasite resistance to AMT.

For the 7G8 lineage, a total loss of CQR was also observed in the back-mutants. CQ IC₅₀ values were 23.9 and 25.5 nM for T76K-1^7G8 and T76K-2^7G8, respectively (Supplementary Table I), which were four- to five-fold lower than C-1^7G8 and C-2^7G8 (P<0.001), and 10- to 11-fold lower than 7G8 (Figure 3). The CQ-sensitive line 3D7 gave a CQ IC₅₀ value of 25.2 nM. Again, the loss was greater with m-dCQ, with T76K-1^7G8 and T76K-2^7G8 showing IC₅₀ values comparable to 3D7 (29–42 nM), 11- to 13-fold lower than C-1^7G8 and C-2^7G8 (P<0.01) and 28- to 29-fold lower than 7G8.

QN assays revealed a statistically significant two-fold decrease in IC₅₀ values in the 7G8 back-mutant lines compared to the controls (P<0.001; Figure 3). Thus, for 7G8 (but not Dd2), K76T appeared to contribute to QN resistance. These different QN results with Dd2 and 7G8 support the idea that pfcrt is but one component of a multifactorial basis of QN resistance (Reed et al, 2000; Sidhu et al, 2002; Ferdig et al, 2004), and that the extent of its contribution differs between strains. For our 7G8 lines, no significant differences were observed for QD, ADQ or MFQ. For AMT, all 7G8 lines displayed high IC₅₀ values (129–213 μM), irrespective of pfcrt codon 76. Thus for 7G8 (in contrast to Dd2), the AMT response appears to be governed either by pfcrt polymorphisms not tested herein or by separate genes.

These assays also revealed a total loss of VP reversibility of CQ and m-dCQ resistance in all Dd2 and 7G8 back-mutant lines (P<0.01; Figures 2 and 3 and Supplementary Table I), implicating residue 76 as a key component of the VP-reversible CQR phenotype in these two geographically distinct parasites. Recombinant control lines retained similar degrees of VP reversibility as parental, nontransfected lines.

Arguably, loss of CQR would inevitably lead to loss of VP reversibility, if there were no longer a mechanism to reverse in CQ-sensitive parasites. Thus, to separate VP reversibility from drug resistance per se, we took advantage of the highly VP-reversible, QN resistance phenotype observed in the Dd2 back-mutants. VP reversibility was quantified using the ‘Response Modification Index' (RMI: defined as the ratio of the IC₅₀ in the presence of VP to that in VP's absence) (Mehlotra et al, 2001). For Dd2 and the controls C-1^Dd2 and C-2^Dd2, the QN RMI was 0.31–0.42 (equivalent to a QN resistance reversal of 69–58%). This was in sharp contrast to T76K-1^Dd2 and T76K-2^Dd2, for which the QN RMI increased to 0.89 and 0.91 (P<0.01; see Supplementary Figure 1). Thus, the loss of T76 abolished VP reversibility of both CQ and QN resistance in Dd2 parasites.

Dd2 and 7G8 are known to differ in their degree of VP reversibility (Mehlotra et al, 2001; Figures 2 and 3), and differ at seven positions in PfCRT, including residues 72, 74 and 75 in transmembrane domain 1 (TMD1) that precede K76T (Table I). To assess whether the preceding TMD1 mutations might be responsible for these differences in VP reversibility, we replaced 7G8-type TMD1 with the corresponding Dd2 sequence, while retaining all downstream 7G8 mutations. Recombinant parasites expressing this chimeric gene were identified and two clones, TMD1-1 and TMD1-2 (Table I), were characterized. Considering amino acids 72–76 as a haplotype, these TMD1 mutants were CVIET, versus CVIEK for the back-mutants (T76K-1^7G8 and T76K-2^7G8) and SVMNT for the recombinant controls (C-1^7G8 and C-2^7G8). Molecular analyses confirmed their recombinant nature (Figure 1; data not shown).

Drug assays (4–7 for each line, performed in duplicate) produced CQ and m-dCQ RMI values, respectively, of 0.37 and 0.43 for TMD1-1 and 0.36 and 0.39 for TMD1-2. In comparison, CQ and m-dCQ RMI values were 0.35 and 0.37 for Dd2 and 0.68 and 0.64 for 7G8 (consistent with reduced VP reversibility in 7G8). CQ RMI values were significantly different between TMD1 lines and 7G8 (P<0.05; Supplementary Figure 1). Yet TMD1 and recombinant controls maintained very similar CQ and m-dCQ IC₅₀ values (Supplementary Table I). TMD1 lines also displayed a statistically significant reduction in QN+VP IC₅₀ values compared to recombinant controls (P<0.05; Figure 3). These data suggest that while residue 76 can largely determine the presence or absence of VP reversibility of CQ or QN resistance, the preceding TMD1 mutations influence the degree of reversibility.

Recombinant lines were assayed (4–5 times in duplicate) for their susceptibility to diaminoalkanes containing the same quinoline ring as CQ yet differing in their side-chain length. Compounds AQ13, AQ26, AQ33 and AQ40 contained, respectively, three, four, six or 12 CH₂ groups, compared to CQ that has a five-carbon side chain (De et al, 1996).

Assays with Dd2 indicated a pronounced bell-shaped curve with highest resistance to CQ (Figure 4). Crossresistance was clearly evident with the analogs that varied by only a single CH₂ group (i.e. AQ26 and AQ33), yet was absent when two CH₂ groups were removed (AQ13) or six were added (AQ40). Results with 7G8 were suggestive of a slightly higher degree of crossresistance to AQ26 and AQ33 (also see Supplementary Table II). Compared to parental lines, recombinant controls displayed lower IC₅₀ values, yet maintained similar crossresistance patterns. Strikingly, removal of K76T ablated this bell-shaped profile in both Dd2 and 7G8, such that resistance was lost to CQ, AQ26 and AQ33.

These analogs were also tested on recombinant GC03 parasites engineered to express Dd2 or 7G8 forms of pfcrt in the place of the wild-type allele (Sidhu et al, 2002). C4^Dd2 and C6^7G8 clearly displayed crossresistance to the most closely related CQ analogs AQ26 and AQ33, closely resembling the respective Dd2 and 7G8 profiles (Figure 4). This pattern was not observed with C2^GC03. Thus, pfcrt mutations appear to largely account for patterns of crossresistance to CQ analogs.

We leveraged the availability of these recombinant lines to further investigate the CQR mechanism, beginning with measurements in infected red blood cells (iRBCs) of saturable CQ accumulation at equilibrium. These measurements can be used to extrapolate apparent affinities of saturable CQ binding (apparent K_d values), which were previously observed to correlate with CQ's antimalarial activity (Bray et al, 1998). Results with parental lines (GC03, Dd2 and 7G8) confirmed a close correlation between apparent K_d (Figure 5A–C) and CQ IC₅₀ values (Supplementary Table I). We then determined whether changes in equilibrium CQ accumulation could be related to PfCRT polymorphisms. Remarkably, the back-mutants displayed a dramatic increase in equilibrium CQ accumulation (resulting in apparent K_d values of 10.0 and 9.2 nM for T76K-1^Dd2 and T76K-1^7G8, as compared to apparent K_d values of 175.6 and 128.3 nM for Dd2 and 7G8, respectively; Figure 5A and B). This paralleled the ability of T76K to ablate CQR in both genetic backgrounds. Recombinant controls displayed CQ accumulation levels that were slightly lower than parental lines.

We also examined the GC03 recombinant lines expressing different pfcrt alleles (Sidhu et al, 2002). Here, the control CQ-sensitive line C2^GC03 displayed CQ accumulation levels equivalent to GC03 (apparent K_d=15.4 and 12.0 nM, respectively). Similarly, the CQ-resistant mutant lines C4^Dd2 and C6^7G8 displayed levels of CQ accumulation (apparent K_d=143.7 and 170.1 nM, respectively) comparable to Dd2 and 7G8 (Figure 5C). Apparent K_d values were highly correlated with CQ IC₅₀ values for these 10 lines (r²=0.91; Figure 5D). These data confirm an association between saturable CQ accumulation at equilibrium and sensitivity to this drug and imply a key role for the PfCRT K76T mutation.

Saturable equilibrium CQ accumulation in iRBCs has been attributed to drug binding to FP (most likely in its dimeric β-hematin state) (Bray et al, 1998). To investigate whether PfCRT might therefore regulate CQ access to FP, we measured [³H]CQ binding to this intracellular target. Results demonstrated a five- to eight-fold greater CQ binding to FP in GC03 as compared to Dd2 and 7G8 (Figure 6A–C). Similarly, T76K-1^Dd2 and T76K-1^7G8, compared with their respective recombinant controls, demonstrated a five-fold increase in CQ–FP binding (P<0.001; Figure 6A and B). Comparing C4^Dd2 or C6^7G8 with C2^GC03, the CQ-resistant mutants displayed a six-fold reduction (P<0.001; Figure 6C). These data suggest that pfcrt mutations largely dictate the amount of CQ–FP binding in iRBCs.

Upon addition of VP, all CQ-resistant parental and recombinant control lines displayed a stimulation of CQ–FP binding (by 2.1- to 2.6-fold and 1.2- to 1.4-fold for lines expressing Dd2 or 7G8 pfcrt alleles, respectively, comparable to changes in IC₅₀ values observed with VP; Supplementary Table I). In contrast, no change was observed with VP for the CQ-sensitive lines T76K-1^Dd2, T76K-1^7G8 and C2^GC03 (Figure 6A–C). This implied that the ability of VP to increase CQ–FP binding was dependent on the presence of PfCRT K76T.

These studies also revealed increased CQ–FP binding in C-1^Dd2 compared to Dd2 (tested±VP), closely mirroring differences between these lines in their CQ±VP IC₅₀ values. However, similar CQ–FP binding was observed for C-1^7G8 and 7G8, despite differences in expression levels and CQ±VP IC₅₀ values. One explanation, aside from possible experimental variability, could be that changes in PfCRT expression levels have a greater impact on CQ–FP binding in Dd2 than in 7G8. Nevertheless, for both genetic backgrounds, K76T appeared equally important in restricting CQ–FP binding (Figure 6A and B). Comparisons for all 10 lines, tested±VP, revealed a strong inverse relationship between CQ–FP binding and CQ IC₅₀ values (r²=0.93; Figure 6D).

For allelic exchange, a 2.9 kb pfcrt fragment containing a 1.6 kb promoter element and 1.3 kb of the gene (exons 1–5) was PCR amplified from 106/1 (Sudan) or Dd2 genomic DNA using primers p1 (ACGGATCCGGTACCTTAGAACCCTAAGAATATCA GCTC; pfcrt 5′UTR-specific; BamHI and KpnI sites underlined) and p2 (TTGCGGCCGCATGCATGTCATGTTTGAAAAGCAT ACAGGC; pfcrt exon 5′-specific; NotI and NsiI sites underlined). Sequence-verified products were subcloned into BamHI–PstI-digested pMini-BSD, which expresses the bsd selectable marker that is under the transcriptional control of calmodulin 5′UTR and hrp2 3′UTR sequences. The resulting constructs harboring the 106/1 and Dd2 pfcrt sequences were termed pK76 and pT76, respectively. Luciferase constructs and assays are described in Supplementary data.

P. falciparum culturing, stable transfection leading to allelic exchange, limiting dilution cloning and 72 h [³H]hypoxanthine incorporation assays (to determine drug IC₅₀ values) were performed as described (Waller et al, 2003). Complete drug response curves obtained from a representative assay are included in Supplementary Figure 2.

The functional downstream recombinant pfcrt locus was detected by PCR using primers p3 (CAATTAACCCTCACTAAAGGG; pBluescript-specific) and p4 (CCCAAGAATAAACATGCGAAACC; pfcrt exon 7-specific) (Figure 1A). The truncated upstream recombinant pfcrt fragment was detected using primers p5 (CTTCAATTCTCATATTTCAATATATTCC; pfcrt 5′UTR-specific) and p6 (GATAGCGATTTTTTTTACTGTCTG; hrp2 3′UTR-specific) (Figure 1A). PCR products were nested using pfcrt primers p7 (AATTCAAGCAAAAATGACGAGCG; exon 1-specific) and p8 (ACTGAACAGGCATCTAACATGG; exon 3-specific). For RT–PCR, cDNA was amplified using primers p7 and p9 (TTCCTACACGGTAAATTATAGAACC; pfcrt exon 12-specific) for the functional gene and primers p7 and p6 for the upstream remnant. RT–PCR products were sequenced across codons 72–76. For Northern blotting, total RNA was resolved, blotted and hybridized with an exon 5–13 cDNA probe generated with primers p10 (CATTTACCATATAATGAAATATGGAC) and p11 (GTTAATTCTCCTTCGGAATCTTCATTTTCTTCAT). For Western blotting, parasites were doubly synchronized, harvested as early-mid trophozoites and protein samples normalized using parasitemia, hematocrit and densitometry (also see Supplementary data).

These were performed as described (Bray et al, 1998). Briefly, synchronized trophozoites were incubated in triplicate for 1 h at 37°C in bicarbonate-free RPMI with 10 mM HEPES, pH 7.4, 1 nM [³H]CQ and 5–250 nM unlabeled CQ. Counts corresponding to nonsaturable CQ uptake into iRBCs (calculated using 100 μM external CQ), as well as CQ uptake by uninfected RBCs, were subtracted from the average total counts to yield saturable CQ uptake at equilibrium. Data were analyzed by nonlinear regression (Marquart method). Standard errors were calculated using matrix inversion (Erithacus Software Ltd, UK).

CQ–FP binding was measured by assessing the incorporation of sublethal concentrations of [³H]CQ into hemozoin crystals (Sullivan et al, 1996). Briefly, synchronized P. falciparum cultures were incubated for 48 h with 1 nM [³H]CQ±0.8 μM VP. Parasites were recovered from saponin-lysed RBCs, washed, resuspended in hypotonic buffer and sonicated. Hemozoin was purified by sucrose centrifugation, washed with SDS, dissolved with NaOH into its monomeric FP form, and FP and [³H]CQ concentrations were determined (see Supplementary data). Earlier studies showed that free FP is required for CQ–hemozoin binding; therefore, these measurements can be used to extrapolate CQ–FP binding (Sullivan et al, 1996, 1998).