In Vitro Efficacy, Resistance Selection, and Structural Modeling Studies Implicate the Malarial Parasite Apicoplast as the Target of Azithromycin*

Azithromycin (AZ), a broad-spectrum antibacterial macrolide that inhibits protein synthesis, also manifests reasonable efficacy as an antimalarial. Its mode of action against malarial parasites, however, has remained undefined. Our in vitro investigations with the human malarial parasite Plasmodium falciparum document a remarkable increase in AZ potency when exposure is prolonged from one to two generations of intraerythrocytic growth, with AZ producing 50% inhibition of parasite growth at concentrations in the mid to low nanomolar range. In our culture-adapted lines, AZ displayed no synergy with chloroquine (CQ), amodiaquine, or artesunate. AZ activity was also unaffected by mutations in the pfcrt (P. falciparum chloroquine resistance transporter) or pfmdr1 (P. falciparum multidrug resistance-1) drug resistance loci, as determined using transgenic lines. We have selected mutant, AZ-resistant 7G8 and Dd2 parasite lines. In the AZ-resistant 7G8 line, the bacterial-like apicoplast large subunit ribosomal RNA harbored a U438C mutation in domain I. Both AZ-resistant lines revealed a G76V mutation in a conserved region of the apicoplast-encoded P. falciparum ribosomal protein L4 (PfRpl4). This protein is predicted to associate with the nuclear genome-encoded P. falciparum ribosomal protein L22 (PfRpl22) and the large subunit rRNA to form the 50 S ribosome polypeptide exit tunnel that can be occupied by AZ. The PfRpl22 sequence remained unchanged. Molecular modeling of mutant PfRpl4 with AZ suggests an altered orientation of the L75 side chain that could preclude AZ binding. These data imply that AZ acts on the apicoplast bacterial-like translation machinery and identify Pfrpl4 as a potential marker of resistance.

Licensed antibiotics with antimalarial activity offer considerable promise as suitable partners for novel combination therapies directed against drug-resistant Plasmodium falciparum infections. Clinical studies have demonstrated good clinical cure rates with treatments that use tetracycline, its more potent derivative doxycycline, or clindamycin, in combination with fast-acting antimalarials such as quinine (1)(2)(3). Whereas the mode of action of these antibiotics has until recently remained elusive for Plasmodium, against bacteria they act by binding to the 70 S ribosome (comprised of the 50 S and 30 S subunits), thereby inhibiting protein synthesis. Tetracycline and doxycycline bind to the 30 S subunit (comprising the 16 S rRNA and ribosomal proteins), whereas clindamycin binds to the 50 S subunit (comprising the 23 S rRNA, the 5 S rRNA, and ribosomal proteins including L4 and L22) (4). The World Health Organization recommends doxycycline as malaria prophylaxis for travelers to endemic areas. Tetracycline and doxycycline, however, manifest toxicity in children below the age of eight and in pregnant women, effectively precluding their use in the most important malaria patient populations (5).
Another antibiotic that appears particularly promising for antimalarial combination therapy is azithromycin (AZ), 2 a 15-membered azalide that has been broadly used for the treatment of streptococcal, staphylococcal, chlamydial, and gonorrheal infections. Notably, AZ displays a good safety profile for young children and pregnant women (6). This semisynthetic second-generation antibiotic was derived from its parent compound erythromycin by introducing a methyl-substituted nitrogen group at position C-9 of the lactone ring, resulting in several improved features. These include a broader spectrum of activity, more favorable pharmacodynamics, and a longer elimination half-life. This extended half-life has been attributed to AZ concentrating inside lysosomes in phagocytic cells, as a result of protonation of its two basic amine groups (7), followed by AZ leaching out into the bloodstream over a period of several days to a week. This provides effective therapeutic coverage with AZ for several days post-completion of treatment.
Against bacteria, AZ acts by binding reversibly to the 50 S ribosomal subunit, in the vicinity of the peptidyltransferase center in domain V and the entrance of the polypeptide tunnel (4). This binding site is formed by the association of domains II and V of the bacterial 23 S rRNA and includes the ribosomal proteins L4 and L22 that line the protein exit tunnel (8 -10). AZ binding interferes with transpeptidation/translocation of nascent polypeptides exiting the polypeptide tunnel, leading to premature detachment of incomplete peptide chains and subsequent cell death (11)(12)(13).
AZ activity against malarial parasites was initially discovered by in vitro testing of chloroquine (CQ)-sensitive and CQ-resistant P. falciparum lines (14,15). Further studies revealed a slow onset of parasite killing, similar to tetracycline, doxycycline, and clindamycin (16). In vitro screening for potential drug combinations was suggestive of additive to synergistic interactions with AZ-CQ as well as AZ-quinine combinations in some CQresistant P. falciparum strains (17). Against CQ-sensitive strains, the effect was additive. In vivo studies, initially conducted in mice infected with blood stage Plasmodium berghei parasites, also showed good treatment efficacy with AZ monotherapy. This efficacy was enhanced by combining AZ with the fast-acting blood schizonticidal agents CQ, quinine, artemisinin, or halofantrine (15,18). AZ curative blood schizonticidal efficacy was also confirmed in Aotus monkeys infected with CQ-resistant P. falciparum (18). Evidence that AZ was clinically effective in treating P. falciparum malaria in humans came from pilot studies showing that AZ monotherapy was an effective prophylactic agent (19,20). A more recent treatment trial in India reported close to 100% cure rates in patients treated with AZ plus CQ, with this combination proving to be far more effective than monotherapy with the individual agents (21). High levels of treatment efficacy were also observed in a recent trial from Thailand that studied regimes of combining AZ with quinine or artesunate (22). A monotherapy trial in India also found AZ to be well tolerated and highly effective in treating Plasmodium vivax malaria (23).
The proven antimalarial efficacy observed with AZ has made it important to define its mode of action against malarial parasites. One pressing question is whether AZ, a weak base, acts inside the parasite digestive vacuole (DV), a lysosomal-like acidic compartment (pH 5.0 -5.4) where CQ and other weak base 4-aminoquinolines concentrate and act by interfering with heme detoxification (24 -26). Such an activity might account for earlier reports of AZ-CQ in vitro synergy (17). Another possibility is that AZ acts by inhibiting the bacterial-like protein synthesis machinery present in the Plasmodium apicoplast, an organelle of cyanobacterial origin that has been shown in the related Apicomplexan parasite Toxoplasma gondii to be indispensable for parasite growth (27)(28)(29). To address these questions, we have investigated the efficacy of AZ alone or in combination with fast-acting antimalarials, the effect on AZ susceptibility of point mutations in DV transmembrane transporters, the selection of AZ-resistant parasite lines, and the molecular mechanisms that can cause AZ resistance. Our results directly implicate the apicoplast 50 S ribosomal subunit as the target of AZ in P. falciparum.

EXPERIMENTAL PROCEDURES
Parasite Culture and in Vitro Antimalarial Drug Assays-P. falciparum parasites were cultured in vitro in human red blood cells as described (30). Antimalarial susceptibilities were measured in 96-well plates over 48 or 96 h of exposure with an initial parasitemia of 0.5 or 0.2% synchronized ring stage parasites, respectively. Drug IC 50 values (producing 50% inhibition of [ 3 H]hypoxanthine uptake) were derived by linear regression from the dose-response curves (30,31). In the 48-h drug assays, parasites were exposed to drug for 30 h, and 0.5 Ci of radiolabeled [ 3 H]hypoxanthine was added to each well for the last 18 h. In the 96-h drug assays, media was replaced after 48 and 72 h, and [ 3 H]hypoxanthine was added at the 72-h time point (0.5 Ci/well). Media replacements contained appropriate amounts of drug to ensure that drug concentrations remained constant in each well throughout the duration of the assay. To test the interaction of AZ with other drugs, each drug was tested alone and in combination at fixed ratios of its IC 50 value, as calculated separately for each assay (AZ, test drug ratios of 0:1, 1:3, 1:5, 1:1, 3:1, 5:1 and 1:0). Fractional IC 50 (FIC 50 ) values from 96-h assays were calculated as previously described (32) and used to plot isobolograms. Two-tailed unpaired Student's t tests were performed for statistical comparisons. AZ dihydrate (CP-62,993-03) was obtained from Pfizer. Thiostrepton, tetracycline, doxycycline, and erythromycin were purchased from Calbiochem, whereas triclosan and CQ were purchased from Sigma.
Selection of AZ-resistant P. falciparum Lines-P. falciparum lines were selected for resistance to AZ using the following regime. First, we exposed Dd2 and 7G8 parasites (5 ϫ 10 10 each) to 38 and 115 nM AZ, respectively (i.e. 30 and 90 ng/ml of AZ dihydrate), for 3 days with twice daily feeding. We then increased the drug concentration to 380 nM for both lines and maintained this level for 4 days. For the following 15 days Dd2 and 7G8 parasite lines were again maintained at 38 and 115 nM AZ, respectively. Giemsa-stained thin blood films were examined every 2 to 3 days to check for viable asexual-stage parasites. Once a week, 30 -40% of the culture was replaced with freshly drawn red blood cells. For both lines, AZ-resistant ring stage parasites were observed by day 21. Once the parasitemias reached 2-3%, parasites were phenotypically characterized for their drug response profiles using [ 3 H]hypoxanthine incorporation assays. Dd2 and 7G8 AZ-resistant parasite lines (termed AZ-R Dd2 and AZ-R 7G8 , respectively) were subsequently cloned by limiting dilution (33) and cultured in the presence of 640 and 1280 nM AZ, respectively.
DNA Constructs and Sequencing-A 1.0-kb DNA fragment encompassing the open reading frame of the P. falciparum apicoplast-encoded ortholog of the rpl4 ribosomal protein gene (denoted Pfrpl4; GenBank TM accession number X95276) was PCR amplified from the wild-type and AZ-resistant parasite lines using primers p1083 (5Ј-AATGGTAACATATCTATTT-TGGGAATAGA) and p1084 (5Ј-AATTTCCACCTCCTT-TATTTAATAGTA). This fragment was used in nested PCR with internal primers p1112 (5Ј-TTatcgatAATATTAT-TATTTTAAATAATAATACA; ClaI site in lowercase) and p1111 (5Ј-AcgtacgAATTAAAAATATTAAATAATTATAC-ATAAT (BsiWI site in lowercase) to amplify a full-length 0.6-kb coding sequence fragment lacking only the ATG start and TAA stop codons. Amplified products were cloned into pCR2.1-TOPO vector (Invitrogen) and sequenced using T7 and M13-reverse primers (Invitrogen).
For the transfection studies, sequence-verified Pfrpl4 inserts were subcloned into the ClaI and BsiWI restriction sites of the expression plasmid pLN-ACP-V5. This plasmid, based on pLN (34), contains a 180-bp sequence encoding the N-terminal bipartite leader sequence of the fatty acid synthase acyl-carrier protein (ACP), used to target proteins to the apicoplast (35), as well as a 3Ј sequence encoding a C-terminal V5 epitope tag (GKPIPNPLLGLDST). The final construct, termed pLN-ACP-Pfrpl4-V5 (supplemental Fig. 1), contained mutant or wild-type Pfrpl4 fused at its 5Ј end to the ACP leader sequence and at its 3Ј end to the V5 tag. This fusion gene is regulated by a 1.0-kb calmodulin 5Ј-untranslated region and a 0.8-kb hsp86 3Ј-untranslated region. Related constructs were made that replaced the 3Ј V5 tag with a GFP-Mut2 (green fluorescent protein) sequence, creating the pLN-ACP-Pfrpl4-GFP plasmid. From these plasmids, we derived pLN-SP-Pfrpl4-V5 and pLN-SP-Pfrpl4-GFP by replacing the full-length ACP apicoplast targeting sequence with a 57-bp signal peptide (SP) sequence from the ACP gene. All of these constructs were also made with a 0.9-kb hsp86 5Ј-untranslated region fragment in place of the calmodulin 5Ј-untranslated region. This resulted in 16 constructs, which were all transfected into Dd2 parasites. The calmodulin promoter series of constructs were also transfected into 3D7 and GC03 parasites. Transfections were carried out as previously described (36) and recombinant lines were selected using 2.5 g/ml blasticidin hydrochloride.
Structural Modeling-Sequences and structures of ribosomal L4 proteins were retrieved from the RCSB Protein Data Bank. Multiple sequence alignments were carried out at the SDSC workbench using CLUSTAL W (37). The model of the PfRpl4 conserved loop from Lys 57 to Pro 97 was built with MODELLER (38). As templates, we used the Escherichia coli and Deinococcus radiodurans L4 proteins, which share 39 and 32% sequence identity with PfRpl4 in this loop region. The AZ molecule in the D. radiodurans ribosome structure (13) also served as a ligand for modeling. The loop of the PfRpl4 G76V mutant was also built using MODELLER with all three wild-type protein structures as templates. The quality of these models was evaluated with DOPE potential (39), and their superimposition and rendering were carried out with PyMol.

AZ Displays Extraordinary Increases in Potency against P. falciparum Strains Exposed for a Second Generation of Growth
as Compared with Thiostrepton, Triclosan, and CQ-A previous study with two P. falciparum lines observed an 80 -220-fold increased potency when the duration of AZ exposure was prolonged from 48 to 96 h (16). We extended this observation to other multidrug-resistant and drug-sensitive lines and compared the effect to that of other antimalarial agents that share similar modes of action or are in clinical use. The lines chosen were Dd2 (Indochina; resistant to CQ, quinine, pyrimethamine, and sulfadoxine), 7G8 (Brazil; resistant to CQ and pyrimethamine), and GC03 (a CQ-sensitive progeny of the HB3 ϫ Dd2 genetic cross (40)). Comparative antimalarial agents included thiostrepton (which targets the GTPase site of the apicoplast LSU rRNA (41,42)); triclosan (a topical antimicrobial that inhibits enoyl-ACP reductase, a component of the fatty acid type II biosynthesis pathway in the Plasmodium apicoplast (43,44)), and CQ (which binds to heme moieties in the parasite DV and prevents their detoxification (45)). A comparative profile of responses to each compound for these parasite lines is presented as IC 50 values in Fig. 1 Table 1). This translated into a substantial 34-, 83-, and 234-fold increase in AZ potency resulting from a second generation of parasite exposure. In comparison, these lines had thiostrepton IC 50 values of 377, 521, and 917 nM, respectively, after 96 h, corresponding to more modest 4.5-, 1.3-, and 2.5-fold increases in susceptibility. In 96-h assays with these same lines, the triclosan IC 50 values were 4186, 424, and 1526 nM, and CQ IC 50 values were 134, 113, and 24 nM, respectively. This corresponded to 1.0 -2.0-fold increases in triclosan or CQ susceptibility. These results indicate that AZ becomes a highly active antimalarial that manifests its potency after a second generation of parasite intraerythrocytic growth. This increase in potency far exceeds other agents that also presumably target apicoplast pathways, including in the case of thiostrepton presumably the same ribosomal subunit involved in protein synthesis.
We also assayed the TM90C6A and TM90C6B Thai isolates that have been more recently adapted to in vitro culture (46). AZ-CQ combination assays performed with these lines also found no evidence for synergy (TM90C6A mean FIC 50 Fig. 2).
In view of the decline of CQ as a first-line antimalarial and the increasing interest in the use of amodiaquine or artesunate as alternative antimalarials (notably for combination therapy), we also assayed AZ-amodiaquine and AZ-artesunate combinations with the drug-resistant Dd2 and 7G8 lines. In 96-h combination assays, mean FIC 50 values for AZ-amodiaquine were 1.28 (95% CI, 1.01-1.56) and 1.23 (95% CI, 0.87-1.58), whereas mean FIC 50 values for AZ-artesunate were 1.62 (95% CI, 1.02-2.22) and 1.34 (95% CI, 1.14 -1.54) against Dd2 and 7G8 parasite lines, respectively (data not shown). Thus, these assays found no evidence of in vitro synergy between AZ and either of these drugs.

Parasite Susceptibility to AZ Is Not Influenced by Mutations in the DV Transmembrane Proteins
PfCRT and PfMDR1-We assessed whether P. falciparum susceptibility to AZ in vitro could be influenced by point mutations in the DV transmembrane proteins PfCRT or PfMDR1 (the latter is also known as Pgh1). This was for several reasons. First, genetic studies have demonstrated that mutant forms of these proteins can affect parasite in vitro susceptibility to a wide range of antimalarials, including CQ, quinine, mefloquine, halofantrine, and the antiinfluenzal agent amantadine (47)(48)(49)(50)(51)(52)(53). Second, there is a published report of AZ-CQ synergism that raised the possibility of their sharing a related mode of action (17). Third, AZ, like CQ, is a weak base that could be predicted to accumulate in the acidic P. falciparum DV (where CQ also accumulates and exerts its antimalarial property (54)). We thus took advantage of existing recombinant lines that had been genetically modified to express mutant forms of either protein (50,52,55) and tested these in 96-h assays with AZ. Results showed AZ IC 50 values in the range of 48 -194 nM for pfcrt or pfmdr1 genetically modified lines, with no significant changes observed between these lines and their parental or recombinant control lines ( Fig.  3; haplotypes and IC 50 values are reported in supplemental Table 2). These results demonstrate that mutations in these DV transporter proteins do not noticeably modulate parasite susceptibility to AZ.
AZ-resistant P. falciparum Lines Can Be Selected in Vitro-To assess whether AZ resistance could be selected in vitro, we subjected Dd2 and 7G8 parasites to moderate levels of AZ (see "Experimental Procedures") and selected resistant lines. These lines, termed AZ-R Dd2 and AZ-R 7G8 , were assayed for their degree of AZ resistance in 96-h assays, in parallel with their parental lines Dd2 and 7G8. This produced AZ IC 50 values of 1.9 and 4.0 M, respectively, as compared with IC 50 values of 124 and 228 nM, respectively, for the parental Dd2 and 7G8 lines, respectively, corresponding to highly significant (p Ͻ 0.001) 16-and 17-fold decreases in AZ susceptibility in the    Table  S3). The fact that resistant lines could be rapidly selected highlights the need to survey for resistance in AZ-containing antimalarial combination therapies.
As noted earlier, AZ and its progenitor erythromycin are chemically very closely related, and cross-resistance between the two antibiotics has been reported in several bacterial species including Streptococcus pneumoniae, Haemophilus influenzae, and Neisseria gonorrhoeae (56 -59). Accordingly, we performed 96-h assays to investigate whether our P. falciparum AZ-resistant mutant lines would also be resistant to erythromycin. Indeed, this was observed but to a smaller extent, with significant (p Ͻ 0.01) 4-and 2-fold increases in erythromycin IC 50 values in AZ-R Dd2 and AZ-R 7G8 , respectively (Fig. 4).
We also screened for possible changes in parasite susceptibility to tetracycline, doxycycline, thiostrepton, and CQ. Tetracycline and doxycycline were chosen because of their clinical use in combination with quinine, primarily for the treatment of uncomplicated, CQ-resistant P. falciparum malaria. As compared with its parental line 7G8, the mutant AZ-R 7G8 line displayed a significant (p Ͻ 0.05) decrease in tetracycline, doxycycline, and thiostrepton IC 50 values, indicating that AZ-R 7G8 had become hypersensitive to these antimalarial drugs. No such change in susceptibility to these drugs was observed in the AZ-R Dd2 line compared with Dd2. The CQ response was unchanged in both mutant AZresistant lines (Fig. 4).
To assess the stability of the AZresistant phenotype, both AZ-selected lines were maintained in the absence of drug pressure for 1 month and the AZ responses then retested. This revealed no alteration in IC 50 values as compared with parasites continuously cultured in the presence of AZ (IC 50 values for AZ-R Dd2 and AZ-R 7G8 were 2.1 and 4.2 M in the absence of drug pressure versus 1.9 and 4.0 M in the presence of sustained drug pressure, respectively, in 96-h assays). These findings indicated that the selected AZ resistance phenotype was stable over time.
Identification of Mutations in the Apicoplast LSU rRNA and Pfrpl4 Genes in AZ-resistant P. falciparum-To examine the molecular basis of AZ resistance, we PCR amplified and sequenced the apicoplast LSU rRNA (the P. falciparum ortholog of bacterial 23 S rRNA) and Pfrpl4 genes as well as the nuclear Pfrpl22 gene (all are present as a single copy). We selected these potential targets based on their similarity with bacterial genes and published reports that mutations in these sequences can confer resistance to AZ and related macrolides in pathogenic bacteria (60 -63).
In examining the P. falciparum apicoplast LSU rRNA (which bears 70% identity to the E. coli 23 S rRNA (60)), we first focused on A1875, which corresponds to the E. coli A2058 nucleotide in the peptidyltransferase center of domain V, and A706, which is equivalent to E. coli A754 in domain II. Mutations in these positions can confer macrolide resistance in bacterial species including E. coli, Helicobacter pylori, S. pneumoniae, and Mycobacterium (62). We also examined G1878 (which corresponds to E. coli G2061) as a mutation at this position can confer resistance to clindamycin and moderately decrease the susceptibility to AZ in T. gondii (64). These nucleotides remained unchanged in our AZ-resistant P. falciparum lines, as did the entire domains II (hairpin loop 35) and V that in , and CQ were determined from 96-h drug assays on AZ-selected mutant lines and parental controls. Values were calculated from five independent assays performed in duplicate. IC 50 values were compared between AZ-R Dd2 and AZ-R 7G8 and their respective parental lines and Mann-Whitney U tests were used to assess for statistical significance (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). Note that the AZ bar chart is represented using a two-segment y axis to cover the data range.
bacteria form a binding pocket for macrolides including AZ and that can also harbor other mutations conferring macrolide resistance (8,12,13,65,66).
Upon sequencing the full-length 2.7-kb apicoplast LSU rRNA gene, we identified a T to C mutation at nucleotide position 438 in the AZ-R 7G8 line. No mutation was observed in this gene in AZ-R Dd2 . Sequence analysis confirmed that both the Dd2 and 7G8 parental lines carried the same T 438 nucleotide, indicating that the AZ-R 7G8 line had acquired this novel LSU rRNA U438C mutation during the period of AZ selection. Intriguingly, this mutation was found in region I, away from domains II and V that contain the majority of the mutations associated with AZ resistance (67).
During this study, we also identified a ribosomal rpl22 ortholog in P. falciparum (Pfrpl22; corresponding to PF14_0642 and originally annotated as encoding a hypothetical protein) based on the similarity of its encoded product with bacterial L22 proteins and an apicolast-directed Rpl22 protein in T. gondii (see sequence alignment in supplemental Fig. 2). Sequence analysis of the full-length Pfrpl22 gene from the AZ-R Dd2 , AZ-R 7G8 , Dd2, and 7G8 parasite lines showed no sequence change.
Analysis of the Pfrpl4 gene, however, identified a single point mutation, resulting in a glycine (Gly, codon GGA) to valine (Val, codon GTA) substitution at codon 76 in both the AZ-R Dd2 and AZ-R 7G8 lines that was not present in either of the parental Dd2 or 7G8 lines. This G76V mutation is present in the conserved P. falciparum motif 71 VQKGLGKAR 79 , which bears close similarity to the bacterial motif 61 RQKGTGRAR 69 (conserved residues underlined) present in the E. coli L4 protein ( Table 1). The fact that both the Dd2 and 7G8 parasite lines were independently selected for AZ resistance and underwent the same mutation in their Pfrpl4 gene provides compelling evidence that this G76V mutation can contribute to P. falciparum resistance to AZ. For both the Pfrpl4 and LSU rRNA genes, the sequences were determined from PCR-amplified products and the electropherograms were unambiguous, implying that the mutant lines had a homogeneous population of mutant apicoplast genomes with no evidence of the wild-type sequences. At this time, we cannot exclude the possibility that undetected mutations occurred elsewhere in the apicoplast or nuclear genomes in our AZ-pressured lines.
Apicoplast-encoded PfRpl4 Cannot Be Redirected to this Organelle from Episomally Replicating Transgene Plasmids-As an initial approach to assess whether the Pfrpl4 G76V mutation was sufficient for AZ resistance, we transfected mutant and wild-type Pfrpl4 alleles into several AZ-sensitive P. falciparum lines (3D7, Dd2, and GCO3). The native Pfrpl4 gene is located on the apicoplast genome, yet our transfection strategies were based on expression of this gene from plasmid DNA that was replicated in the nucleus. In an attempt to target recombinant PfRpl4 to its host apicoplast organelle, these alleles were ligated to a DNA fragment encoding the complete apicoplast targeting sequence of ACP as this sequence was previously shown to direct trafficking of nuclear-encoded gene products to the apicoplast (35). This bipartite targeting sequence is composed of a SP and a transit peptide, and the apicoplast targeting sequence algorithm PlasmoAP (68) predicted the presence of a putative transit peptide sequence in PfRpl4. Accordingly, we also generated constructs that fused just the SP sequence from the ACP gene to mutant or wild-type Pfrpl4. The recombinant ACP-Pfrpl4 and SP-Pfrpl4 mutant or wild-type gene fusions (with a V5 epitope tag or GFP at the 3Ј end) were cloned into expression plasmids under the control of calmodulin or hsp86 promoters, making a total of 16 constructs (69). Parasite lines transfected with sequence-verified plasmids expressing ACP- Azithromycin Acts upon the P. falciparum Apicoplast JANUARY 26, 2007 • VOLUME 282 • NUMBER 4 Pfrpl4-GFP or ACP-Pfrpl4-V5 revealed no detectable protein expression (data not shown). Lines transfected with plasmids expressing SP-Pfrpl4-GFP fusions yielded cultures that expressed recombinant PfRpl4-GFP fusions. However, fluorescence microscopy revealed cytosolic localization of these fusions proteins (supplemental Fig. 1), indicating that the combined SP and Pfrpl4 putative transit peptide sequence were insufficient for apicoplast targeting. These experiments were repeated on several occasions with the same outcomes, independent of which promoter (calmodulin or hsp86) was used to drive the transgene or whether the Pfrpl4 sequence was mutant or wild-type. We speculate that the inability to traffic PfRpl4 to the apicoplast might reflect tight regulation of the apicoplast ribosomal translation machinery that is incompatible with PfRpl4 overexpression in transfected parasites.

Modeling of Mutant PfRpl4 Supports a Role for G76V in AZ
Resistance-Recent elucidation of the crystal structure of the large ribosomal subunit (50 S) of D. radiodurans has provided evidence favoring simultaneous binding of two AZ molecules in the peptide exit tunnel of the 50 S ribosomal subunit (13). One AZ molecule (AZ-1) interacted exclusively with domains IV and V of the 23 S rRNA, whereas a second molecule (AZ-2) interacted with the ribosomal proteins L4 and L22 and the 23 S rRNA domain II. To date, AZ is the only macrolide reported to interact directly with these ribosomal proteins. The major source of interactions with AZ-2 involved the formation of hydrogen bonds with residues Gly 60 and Thr 64 in the D. radiodurans L4 loop (corresponding to residues Val 71 and Leu 75 in PfRpl4; Table 1). This loop (Arg 46 -Pro 86 ) shares 32% identity between D. radiodurans and the corresponding loop (Lys 57 -Pro 97 ) in PfRpl4. Using the D. radiodurans structural data, we modeled the mutant and wild-type PfRpl4 loop as a complex with AZ (Fig. 5). The wild-type PfRpl4:AZ molecular model is supportive of AZ interactions with PfRpl4 residues Val 71 and Leu 75 (Fig. 5A). Superimposition of the mutant PfRpl4 loop structure predicted that the proximal G76V mutation causes a conformational shift in the side chain of Leu 75 , which would be expected to sterically interfere with the AZ-2 binding pocket (Fig. 5B). Of all the possible rotamer positions of the 76V side chain, none were predicted to allow the Leu 75 residue to stay in the wild-type conformation. Thus, these structural models suggest that the G76V mutation causes a conformational change in the neighboring Leu 75 residue that creates spatial conflict with AZ binding to this molecule.

DISCUSSION
Our investigations into the antimalarial properties of AZ reveal this to be a slow-acting drug that becomes highly potent when extending parasite exposure from one to two generations, attaining IC 50 values as low as 50 nM (supplemental Table 2). AZ activity was not synergistic with CQ, amodiaquine, or artesunate. Furthermore, AZ potency was unaffected by point mutations in the DV transmembrane proteins PfCRT or FIGURE 5. Structural models of the PfRpl4 wild-type (A) and G76V mutant (B) complexed with AZ. These models, derived using the D. radiodurans crystal structure of the L4 protein as a template (13), predict a steric clash between the side chain of Leu 75 in the G76V mutant and AZ, consistent with this mutation conferring AZ resistance.
PfMDR1 that can reduce susceptibility to several antimalarials that act inside the DV, suggesting that this is not the primary site of action of AZ. Drug pressure studies resulted in the selection of AZ-resistant parasite lines that were found to harbor point mutations in the apicoplast-encoded Pfrpl4 and LSU rRNA genes involved in apicoplast protein synthesis. These AZ-resistant parasites also displayed cross-resistance to erythromycin, a known inhibitor of the bacterial 50 S ribosomal subunit (56, 57, 70 -73). Collectively, these data provide compelling evidence that the antimalarial properties of AZ are a result of its binding to the apicoplast 50 S ribosomal subunit and inhibiting protein synthesis in this organelle. These data complement separate studies directly implicating the apicoplast as the target of tetracycline and doxycycline in P. falciparum and clindamycin in T. gondii (64,74).
In our study, AZ potency increased 30 -230-fold upon prolonged exposure (from 48 to 96 h). These data are consistent with an earlier study (16) and provide a marked contrast to the comparator drugs thiostrepton and triclosan (that are both also thought to act in the apicoplast (41,42,44,75)) and CQ (that acts in the DV (54)), which showed only minor gains in potency (Fig. 1). This "delayed death" phenotype was first described with T. gondii tachyzoites exposed to clindamycin and AZ (76). These antibiotics were found to not affect replication in the first generation of exposure, yet drastically reduced tachyzoite division in the second generation (64,76). One proposed explanation for this phenotype was that these antibiotics render the parasites unable to establish a parasitophorous vacuole upon host cell reinvasion (28,76). Alternatively, their effect on apicoplast protein synthesis might lead to insufficient levels of an as yet unidentified, apicoplast-encoded parasite factor required for the successful development of the progeny of drug-treated parasites (29,74,77).
Our in vitro studies reveal essentially additive effects between AZ and CQ, which is in contrast to earlier reports of in vitro synergy between these drugs with some P. falciparum lines (17). This discrepancy may be attributable to differences in methodology or analysis, or differences between the lines employed and for how long they had been adapted to in vitro culture. Our lack of synergy, reported with five lines tested on five separate occasions, provides evidence that these drugs have separate modes of action. AZ and CQ may nevertheless manifest some synergistic effects when used clinically. This is evidenced by a recent clinical trial in India in an area with a high prevalence of CQ-resistant malaria, which found that AZ-CQ combination therapy was significantly more efficacious in treating falciparum malaria than monotherapy with either agent (97% cure rates with AZ-CQ versus 33 and 27% with these individual agents, respectively (21)).
Direct evidence that AZ resistance in P. falciparum is mediated by changes in the bacterial-like apicoplast ribosome comes from our resistant mutant in vitro selection studies, which identified mutations in the apicoplast-encoded Pfrpl4 and LSU rRNA genes. A key role for Pfrpl4 is suggested by our finding that both the drug-pressured 7G8 and Dd2 lines underwent a G76V mutation in this gene. Similarly placed mutations in genes encoding L4 have been strongly implicated in macrolide resistance in several bacterial species (Table 1). For example, two studies observed G69C or G71R mutations in this gene in S. pneumoniae strains selected for AZ resistance (56,57). These mutations correspond to residues 74 and 76 in PfRpl4 (Table 1). Other AZ-resistant S. pneumoniae strains also show insertions, deletions, or multiple mutations in this L4 conserved site (71,73). Similar L4 mutations have also been identified in macrolide-resistant strains of Streptococcus pyogenes, H. influenzae, Staphylococcus aureus, and E. coli (70,78,79). Whereas many of those studies only reported an association, genetic complementation studies have provided convincing evidence that L4 mutations can confer AZ resistance. For example, genetic transformation of the AZ-sensitive S. pneumoniae R6 strain with the mutant rpl4 alleles G69C, 67 QKG 69 to QSQKG, 69 GTG 71 to TPS, 67 QKGT 70 to QT, and 64 PWRQ 67 to PQ produced 8-, 15-, 60-, 16-, and 16-fold decreases, respectively, in the susceptibility of these transformed strains to AZ (56, 80 -82). In the converse approach, complementation of erythromycin-resistant E. coli mutants with plasmids expressing wild-type rpl4 or rpl22 genes resulted in decreased resistance to erythromycin (70). Our study also adopted a complementation approach by trying to express the apicoplast-encoded Pfrpl4 gene (both wild and mutant types) in different AZ-sensitive parasite strains. However, despite multiple and varied attempts, we could not obtain recombinant parasite lines that targeted recombinant PfRpl4 to the apicoplast. An alternative and more definitive approach would be to perform allelic exchange of the endogenous Pfrpl4 gene with a mutant Pfrpl4 sequence that is being replicated inside the apicoplast. This experiment can only be undertaken once there is a suitable selectable marker for transfection of the P. falciparum apicoplast.
We note that one of our AZ-resistant lines, AZ-R 7G8 , also harbored a LSU rRNA mutation (U438C) in domain I. This is away from domains II and V that form part of the macrolide binding pocket in the bacterial 50 S ribosomal subunit and that house the majority of the 23 S rRNA mutations associated with AZ resistance (67). Nonetheless, recent reports document the presence of 23 S rRNA domain I mutations (A138G, A260G, A373T, T389C, or A449C) in macrolide-resistant clinical isolates of S. pneumoniae (83,84). This might contribute to the fact that our AZ-R 7G8 mutant had a higher AZ IC 50 value than AZ-R Dd2 . The AZ-R 7G8 line also displayed significantly increased in vitro sensitivity to thiostrepton, tetracycline, and doxycycline. Thiostrepton, a peptide antibiotic, inhibits protein translation by blocking GTPase activity in the 23 S rRNA domain II (85), and studies have shown that an A1067U mutation (E. coli numbering) can confer thiostrepton resistance in E. coli and P. falciparum (42,86). Tetracycline and doxycycline block protein synthesis by binding to the 30 S subunit that inhibits attachment of tRNA to the A-site of ribosomes (87,88). Photolysis experiments performed with radiolabeled tetracycline and E. coli ribosomes, however, provide evidence that this antibiotic can also interact with nucleotides in the central loop of the 23 S rRNA domain V (89). Domains II and V, however, were invariant in AZ-R 7G8 , suggesting that the LSU rRNA domain I mutation we observed in this line was responsible for the observed hypersensitivity to thiostrepton, tetracycline, and doxycycline. A possible explanation is provided by structural studies. Analysis of the structure of the large ribosomal subunit from Haloarcula marismortui complexed with its repertoire of ribosomal proteins showed that the three-dimensional catalytic structure of the 23 S rRNA is maintained by interactions of all six domain subunits with each other through extensive hydrogen bonding and other non-covalent interactions (67). Cryoelectron microscopy studies have also shown that mutant L4 proteins can exert long-range structural effects on the 30 S subunit in erythromycin-resistant E. coli ribosomes (10). We posit that the hypersensitivity of AZ-R 7G8 to thiostrepton, tetracycline, and doxycycline might be due to the LSU rRNA domain I U438C mutation causing structural changes that are transmitted to the antibiotic binding sites on the 50 S and 30 S rRNA subunits. We also note that this increased antibiotic sensitivity, if found to occur frequently in AZ-resistant parasites, would promote the potential use of these antibiotics as partner drugs for combination therapy. Our findings of mutations in pfrpl4 and the LSU rRNA are consistent with published data from pathogenic bacteria (including S. pneumoniae and H. influenzae) that directly implicate mutations in the 23 S rRNA or the genes encoding ribosomal proteins L4 and L22 as causal determinants of macrolide resistance (62,79,90). Furthermore, our finding that resistance to AZ could be rapidly selected in vitro is an important consideration in the testing of future AZ-containing antimalarial combination therapies.
Other reported mechanisms of macrolide resistance in bacteria are methylation of the A2058 nucleotide in the 23 S rRNA domain V by a methyltransferase encoded by the erm (erythromycin ribosome methylation) genes, and drug efflux mediated by an efflux pump encoded by the mef (macrolide efflux) genes (62,92). Methylation of this critical nucleotide acts to reduce macrolide binding to the ribosome, and tends to predominate over point mutations as the major mechanism of macrolide resistance in bacterial species such as Streptococcus and Staphylococcus that harbor multiple rRNA operons. Our analysis of the recently released P. falciparum (HB3 strain) apicoplast genome (GenBank accession number DQ642846, sequenced by the Broad Institute, Cambridge, MA) reveals that this 29-kb genome has only one functional copy of the LSU and small subunit rRNA genes. This resolves earlier questions about the possible presence of a second LSU gene in the apicoplast (61). Our analysis also shows the absence of any putative apicoplast gene coding for an adenosine methyltransferase or an active drug efflux pump. Thus, at this time it would appear less likely that rRNA methylation or drug efflux mechanisms could operate to confer AZ resistance in P. falciparum.
L4 and L22 are the early assembly proteins of the 50 S ribosomal subunit and play a pivotal role in scaffolding the ribosomal domains (10). The crystal structure of L4 from Thermotoga maritima defines a loop of 55 residues that displays the highest degree of phylogenetic conservation across the L4 sequences (corresponding in PfRpl4 to residues 47-101) and that represents the main rRNA binding site (93). RNA-protein cross-linking studies in E. coli have shown that this conserved L4 loop interacts with 23 S rRNA nucleotides 284 -350 and 580 -670 in domains I and II, respectively (94). Cryo-electron microscopy of erythromycin-resistant E. coli ribosomes indicates that sequence changes in L4 or L22 can alter the spatial organization of nucleotides in 23 S rRNA domains II, III, and V, and the conformation of the C-terminal ends of these proteins that form a segment of the polypeptide exit tunnel. These structural rearrangements can significantly change the diameter of the peptide tunnel, affecting macrolide accessibility to the peptidyltransferase center in domain V and preventing the macrolide from inhibiting protein synthesis and egress from the peptide tunnel (10,91). Structural analysis of the entire 50 S ribosomal subunit of H. marismortui containing L4 and L22 has localized binding of AZ and its progenitor erythromycin to a hydrophobic drug-binding pocket formed by critical 23 S nucleotides A 2058 , A 2059 , and C 2611 (E. coli numbering). One key interaction is the formation of a hydrogen bond between the desosamine sugar at position C-5 of the macrolide lactone ring and A 2058 (65). A similar pattern of interactions between AZ and these rRNA nucleotides was also identified in crystallography studies of the D. radiodurans large subunit ribosome. This work also identified an additional binding site for AZ, formed by ribosomal proteins L4 and L22 and domain II of the 23 S rRNA (13). AZ was observed to directly interact with ribosomal proteins L4 (notably with the Tyr 59 , Gly 60 , Gly 63 , and Thr 64 residues) and L22 (the Arg 111 residue) through hydrogen bonding and hydrophobic interactions, respectively (13).
Our modeling of wild-type and mutant PfRpl4, based on structural data from D. radiodurans that identified AZ binding to L4 (13), predicts a dramatic impact of the G76V mutation on the neighboring Leu 75 residue as a result of its reorientation in the mutant protein. This leads us to propose that the reoriented Leu 75 residue can create a steric clash with AZ, resulting in reduced binding. An alternative explanation is that this G76V mutation prevents AZ from binding to the nascent polypeptide exit tunnel in the 50 S ribosome by constricting the macrolidebinding site. In either event, this study provides compelling evidence that AZ acts upon the P. falciparum apicoplast and highlights the importance of screening for possible mutations in the LSU rRNA, Pfrpl4 and Pfrpl22 genes in ongoing clinical trials designed to assess the efficacy of AZ-containing antimalarial combination therapies.