Artemisinin resistance in rodent malaria - mutation in the AP2 adaptor μ-chain suggests involvement of endocytosis and membrane protein trafficking

Background The control of malaria, caused by Plasmodium falciparum, is hampered by the relentless evolution of drug resistance. Because artemisinin derivatives are now used in the most effective anti-malarial therapy, resistance to artemisinin would be catastrophic. Indeed, studies suggest that artemisinin resistance has already appeared in natural infections. Understanding the mechanisms of resistance would help to prolong the effective lifetime of these drugs. Genetic markers of resistance are therefore required urgently. Previously, a mutation in a de-ubiquitinating enzyme was shown to confer artemisinin resistance in the rodent malaria parasite Plasmodium chabaudi. Methods Here, for a mutant P. chabaudi malaria parasite and its immediate progenitor, the in vivo artemisinin resistance phenotypes and the mutations arising using Illumina whole-genome re-sequencing were compared. Results An increased artemisinin resistance phenotype is accompanied by one non-synonymous substitution. The mutated gene encodes the μ-chain of the AP2 adaptor complex, a component of the endocytic machinery. Homology models indicate that the mutated residue interacts with a cargo recognition sequence. In natural infections of the human malaria parasite P. falciparum, 12 polymorphisms (nine SNPs and three indels) were identified in the orthologous gene. Conclusion An increased artemisinin-resistant phenotype occurs along with a mutation in a functional element of the AP2 adaptor protein complex. This suggests that endocytosis and trafficking of membrane proteins may be involved, generating new insights into possible mechanisms of resistance. The genotypes of this adaptor protein can be evaluated for its role in artemisinin responses in human infections of P. falciparum.


Background
Without an effective vaccine, prevention and treatment of human malaria has traditionally relied on chemoprophylaxis and/or chemotherapy [1]. However, Plasmodium falciparum has developed resistance to nearly every anti-malarial drug introduced to date, compromising its control. Resistance arises via the selection of parasites bearing specific mutations, and is decisive in determining the effective lifetime of anti-malarial agents.
Artemisinin combination therapy (ACT) is now a widely used anti-malarial treatment. The artemisinin component is a highly effective and rapidly acting drug [2]. The understanding of its mode of action is incomplete [3,4], but independent lines of evidence from a number of laboratories suggest that its action depends upon its endoperoxide group and activation by haem or other iron sources [3]. Downstream, it may localize close to the digestive vacuole (DV) [5] and effect changes in DV morphology [6] or the distribution, endocytosis and digestion of haemoglobin [7].
Alarmingly, recent data indicate that resistance to artesunate, one of several artemisinin derivatives (ARTDs), is emerging in Cambodia and Thailand [9][10][11][12][13][14]. Molecular (DNA) markers of resistance are therefore required urgently. These tools will help to monitor the evolution of resistance, to establish rational treatment policies and to design drug combinations that delay the evolution of resistance. At present however, in P. falciparum, there are no universally accepted and validated molecular markers of artemisinin resistance in the field. Polymorphisms in PfATPase6 [15,16] or amplification of the multidrug resistance gene, Pfmdr1 [17][18][19] have been investigated. However, no correlation was found between variants of these genes and in vivo responses to ARTDs in the first suggested cases of resistance along the Thai/Cambodian border [20,21].
Candidate gene approaches may not focus on the critical genes nor grasp the full complexity of the drug response mechanisms. Instead, analysis of genetic haplotype variation and conservation and geographic differentiation using genome-wide SNP typing of P. falciparum parasites from Southeast Asia (Thailand, Cambodia and Laos), where resistance to ARTDs is emerging, has identified candidate regions associated with slow parasite clearance rates after drug treatment [21], particularly in a region on chromosome 13. No specific genes in this region were identified as candidates for influencing artemisinin response.
Other experimental studies that do not prejudge the critical genes are especially informative; for example, the in vitro generation of mutant P. falciparum parasites resistant to drugs. Parasites resistant to ARTs have been generated and mdr1 duplications identified [22], but often the phenotypes and genetic changes have tended to be unstable in the absence of drugs [23,24]. Genetic linkage analysis of the experimental P. falciparum Hb3 x Dd2 cross has shown that three loci, including pfmdr1 and two additional loci (on chromosomes 12 and 13) [24] that were associated with artemisinin (ART) responses or the potential to evolve ART resistance, but neither of these parental parasites offered a distinct ART-R phenotype.
In vivo experimental studies using the rodent malaria Plasmodium chabaudi can circumvent some limitations faced by in vitro experimentation. A lineage (strain AS) of genetically stable mutant parasite clones that are resistant to various drugs has been generated by experimental evolution under drug selection ( Figure 1). The genetic mutations conferring resistance to pyrimethamine, sulphadoxine, chloroquine, mefloquine, lumefantrine and artemisinin have been mapped using efficient genetic linkage mapping (population based genome-wide scans of selection) and Illumina whole-genome sequencing [25][26][27][28]. The identification of the critical mutations exploits four important features relevant to the present study.
Firstly, the resistance phenotype appears at the same time/position in the lineage as the critical mutation. Secondly, the mutation will lie at the bottom of a 'selection valley' (genomic region selected by drug). Thirdly, these studies exploited a previously completed reference genome sequence isogenic to the progenitor parasite, AS-sens. Fourthly, the numbers of point mutations (genome-wide) fixed at each step by selection or during cloning are very small, typically 1-3 [26].
For artemisinin, a mutation in a de-ubiquitinating enzyme (V2728F ubp1) was proposed previously as the critical determinant of an artemisinin-resistance (ART-R) phenotype [25] in parasite AS-30CQ ( Figure 1). This ubp1 mutation has also been shown to contribute to resistance to mefloquine and lumefantrine [27] and higher doses of chloroquine [26], as well as ARTs, suggesting that ubp1 may mediate parasite responses to multiple drugs, as does mdr1 duplication [27].
Here, another parasite, AS-ART ( Figure 1), derived from AS-30CQ after prolonged and progressive ART selection [29], is investigated. An increased ART-R phenotype is characterized and a single point mutation in its genome is defined. The 3D structures of homologues of the mutated protein are investigated for possible functional consequences. Allelic variations of the P. falciparum orthologue of the mutated gene are defined in a set of P. falciparum field samples. All the parasite clones used in this study are members of the AS lineage ( Figure 1). Three parasite clones were used: i) AS-3CQ, a parasite that is resistant to low levels of chloroquine [34]; ii) AS-30CQ, that is resistant to high levels of chloroquine [34]; and, iii) AS-ART, which was derived from AS-30CQ through multiple sub-inoculations in mice under increasing doses of artemisinin (ART) [29]. All parasites were routinely inoculated, passaged in CBA mice (four to six weeks) and cryopreserved as previously described [30]. Parasites were prepared and DNA extracted as previously described [35], ensuring that host white cells were removed by CF11 cellulose (Whatman) and Plasmodipur filters (Eurodiagnostica). DNA samples from other clones within the lineage were used to determine the earliest appearance of specific mutations within the lineage. All animal work was conducted according to relevant national and international guidelines: in Portugal, after approval by the Ethics Committee of the Instituto de Higiene e Medicina Tropical of Lisbon, Portugal, under PARECER 2/2006 from August 1st 2006 and in the UK, in compliance with the UK Animals (Scientific Procedures) Act 1986.

In vivo ART drug tests
In vivo ART response phenotypes were assessed as follows. Four to six-week-old inbred CBA mice were divided into nine groups of three mice each. Mice in groups 1-3 were inoculated with 10exp6 parasitized red blood cells (pRBC) of AS-3CQ, AS-30CQ or AS-ART, respectively and given the diluting solvent DMSO orally (untreated controls). Treatment groups 4-6 were inoculated with 10exp6 pRBC of either AS-3CQ, AS-30CQ or AS-ART and treated with an oral daily dose of 200 mg ART kg -1 mouse bodyweight, administered for a total of three consecutive days, starting day 1 post-inoculum (pi). Treatment groups 7-9 were inoculated with 10exp7 pRBC of either AS-3CQ, AS-30CQ or AS-ART and treated with 200 mg ART kg -1 mouse bodyweight, administered for a total of five consecutive days. Individual percentage (%) parasitaemia was followed from day 4 pi onwards and up to day 15 in the case of untreated control mice. For treated mice, % parasitaemia was first assessed one day after the last day of treatment and up to day 18 pi. Results were expressed as daily average % parasitaemia ± standard error from the three mice within each experimental group.

Genome-wide resequencing
Clone AS-sens was previously resequenced with the Illumina W platform using 36 bp single reads [25]. Clone AS-ART was also re-sequenced using 36 bp single-end reads. Individual sequence strings (reads) from AS-sens and AS-ART were aligned against the isogenic AS reference genome assembly (AS-WTSI [36], provided by the Wellcome Trust Sanger Institute, using two different software packages: MAQ (Mapping and Assembly with Quality) [37] and SSAHA2 (Sequence Search and Alignment by Hashing Algorithm) [38]. The AS-WTSI sequence data consisted of a recently completed assembly and annotation made available during this investigation. Detection of SNPs was performed with Samtools [39] and MAQ, using default parameters, as described in Hunt et al. [25]. Small (≤3 bp) indels were detected using Samtools internal algorithm only. The list of single nucleotide polymorphisms (SNPs) and small indels were further filtered by removing mutations proposed for both AS-sens and AS-ART and that therefore did not arise within the AS lineage. Heterozygous and "multiple variant" SNPs, as well as small indels called by less than three reads and less than 50% of the total reads were also removed. Larger indels (>3 bp) and CNVs were detected with both Figure 1 The AS lineage of drug resistant parasites, featuring artemisinin (ART) phenotypes and ubp-1 genotypes. Parasites were selected by passage in presence of drugs shown [29][30][31][32][33]. Artemisinin resistance (phenotype 1) appears during selection by chloroquine [25,26] and is mediated by either of two mutations in the ubp-1 gene (PCHAS_020720) [25,27]. AS-ART was generated during selection by artemisinin [29]. Its increased resistance to artemisinin (phenotype 2) and its genotype is the subject of the present study. *, parasites used in the present study; wt, wild-type; kg bw, kilograms body weight.
MAQ and SSAHA2 using "comparative coverage" analysis, which measures the ratio of the relative coverage (local coverage divided by overall mean coverage) in AS-ART relative to AS-sens. A comparative coverage ratio >1.5 over 200 bp for defining CNVs and <0.25 over 10 bp for defining indels (both adjusted for different genome coverage in AS-sens and AS-ART) were used, as previously described [25]. Unlike previous work, "comparative coverage" analysis was also performed using MAQ. This required running the "pileup" command on both the AS-ART and AS-sens reads using the variable "-q 1" (which excludes reads with a mapping quality <1). The pileup files thus produced were then used for "comparative coverage" analysis with custom made scripts, similarly to the SSAHA2 approach. Confidence levels were assigned to mutations based on the following criteria: a) Samtools quality scores (for SNPs) and b) identification by both MAQ and Samtools (for SNPs and large indels/CNVs). All high confidence putative point mutations proposed by MAQ and/or Samtools were verified by di-deoxy sequencing. Only limited verification was done for potential indels and CNVs. Small indels were defined as "low confidence" mutations by default, due to a majority of calls being confirmed as false positives by di-deoxy sequencing in this and other studies [25,28].

Protein structure-function analysis
The P. chabaudi AP2-μ sequence was used to search the protein structure database (PDB) for homologues. The closest match found was the μ chain of AP2 from Rattus norvegicus, with a bit score of 167 and an E-value of 2e-41. 11 different structure entries of this and related proteins and complexes were present in the PDB. Structure 2PR9 was selected because of the level of R-factors (R: 0.204, R-free: 0.240) and its resolution (2.51Å).
The I-TASSER server [40] was used to construct a homology model of the P. chabaudi homologue, with the R. norvegicus 2PR9 structure as a template onto which the P. chabaudi sequence was modelled. The same method was used to construct a P. falciparum μ model.

Plasmodium falciparum ap2-μ genotyping
In order to investigate the genetic variation in the P. falciparum ap2-μ gene in natural infections, DNA was extracted from field samples collected in Africa and South America, as well as from the reference strain 3D7, according to the method described by Plowe and coworkers [41]. The P. falciparum ap2-μ gene was thus sequenced in a subset of 24 P. falciparum isolates collected within the scope of previous studies, of which eight were from Rwanda [42], eight from São Tomé [43] and eight from Brazil [44].
The sequence of the P. falciparum orthologue of the ap2-μ gene (accession no. PF3D7_1218300) was retrieved from PlasmoDB and used as template for designing primers to amplify its open reading frame using an overlapping PCR fragment strategy (see Additional file 1). PCR assays were performed with 1 μl of DNA into a 50 μl mixture containing 0.2 μM of each primer, 1.5 mM MgCl2, 0.2 mM deoxynucleoside triphosphates and 1.25 units of GoTaq W Flexi DNA Polymerase (Promega). PCR products were analysed by ethidium bromide-stained agarose 2% gel electrophoresis and sequenced directly in both sense and antisense directions with the appropriate PCR amplification primers.

Results
The Plasmodium chabaudi AS-ART phenotype Here, the ART responses of isogenic parasites of the P. chabaudi AS lineage are investigated further; namely, AS-3CQ (ART-S clone), AS-30CQ (ART-R clone) and the AS-ART (ART-R clone) generated during selection by ART ( Figure 1) [29]. In previous work, artemisinin resistance at 100 mg ART kg -1 (body weight) d -1 was demonstrated in both AS-30CQ and AS-ART relative to other clones, which preceded them in the AS lineage, including AS-3CQ [25]. Here, the hypothesis that AS-ART would resist higher doses of ART than AS-30CQ is tested. Parasites were treated for three (1 × 10 exp6 inoculum) or five (1 × 10 exp7 inoculum) days (treatments A and B, respectively) with a daily dose of 200 mg ART kg -1 d -1 and their response to the drug was assessed by comparing both peak parasitaemia and the time taken for each parasite to recrudesce after treatment.
In the absence of treatment, all three parasite clones produced their peak parasitaemia of between 23% (AS-ART) and 40% (AS-30CQ), appearing between days 8 and 9 pi ( Figure 2A). Under either ART treatment A and B however, no parasites were detected in mice infected with AS-3CQ over the follow-up period ( Figure 2B and 2C). After treatment A, AS-30CQ produced detectable parasites between days 13-14, with parasitaemia reaching their peak (~12%) on day 17. However, mice infected with AS-ART presented more rapid recrudescence (days 8-9), with parasitaemia reaching~12% on day 15 after treatment ( Figure 2B). A similar trend was observed after treatment B: mice infected with AS-30CQ first presented a detectable parasitaemia between days 13-14 and peak parasitaemia (~10% on day 18), whilst the AS-ART group recrudesced three days earlier ( Figure 2C) achieving a similar peak parasitaemia on day 15.
As previously reported [25] it is concluded that AS-3CQ is sensitive to ART treatment whereas both AS-30CQ and AS-ART are resistant. Here, however, at high ART doses, AS-ART shows both earlier recrudescence and reduced time to reach peak parasitaemia relative to AS-30CQ. It is concluded that AS-ART has a higher degree of resistance to ART. Here, this enhanced resistance phenotype is called ART-R phenotype 2. The response of AS-30CQ  All point mutations and 34 bp deletion (*) were confirmed by di-deoxy sequencing. A small deletion on chromosome 4 and large deletions on chromosome 5 and 'bin' (**), have strong supporting evidence (extended area of low and zero coverage, SNP-proxies for deletions). Base quality scores (which only apply to SNPs) are derived from the SSAHA2 algorithm (with 99 being the best score). For indels, the approximate size of the affected area (as measured by SSAHA2) and a comparative coverage are given. Amino acid substitutions (non-synonymous mutations) are in bold characters. One point mutation in AP2-μ chain specific to AS-ART (relative to preceding clone, AS-30CQ) is in bold and underline characteres. Tbc: to be confirmed. and AS-ART to 100 mg ART kg -1 d -1 (3 day) is called ART-R phenotype 1.
Genome re-sequencing of AS-ART were mapped in unique positions. Some 98% of the nucleotides in the genome were covered by at least 10 reads and 93% by at least 40 reads; 1.67% was covered by less than three reads (the minimum required for SNP and small indel detection). Similar data were obtained using an alternative mapping software, MAQ. Clones AS-sens and AS-30CQ had been re-sequenced previously and reads aligned similarly [25,26,28].
Two point mutations have not been linked to any particular phenotype. These are: Y162H (quality score 99) in a gene annotated as a hypothetical protein (PCHAS_101550) appearing first in AS-30CQ, and an intergenic mutation (quality score: 92) at base position 936,945 on chromosome 14, located between genes PCHAS_142590 and PCHAS_142600 (both hypothetical proteins with no significant Pfam matches), first appearing in AS-PYR1, the first drug-resistance clone of the lineage.
One point mutation only in clone AS-ART relative to AS-30CQ was confirmed. This mutation was I568T (quality score 99) in a gene (PCHAS_143590) predicted to encode the mu (μ) chain of the AP2 adaptor protein complex, which, in other organisms, is involved with clathrin-mediated endocytosis [45]. This gene was denoted ap2-μ.
In contrast to the eight confirmed mutations above, the other 11 low-quality calls (see Additional file 3, highlighted in orange) were predicted only by Samtools; nine of these have extremely low Samtools quality scores (2 to 13) and were variously mapped to the extreme ends of chrs 01, 03, 07 (two), 10 and 12, where P. chabaudi-specific genes are located (often in multigene families), or to contigs yet to be mapped to the final genome assembly (three in 'bin'). Here, read alignment may be less reliable, as reflected in their low mapping quality. A similar low quality SNP call in AS-30CQ was previously identified as a false positive by di-deoxy-sequencing [28]. Accordingly, these nine low confidence SNPs are strongly predicted to be false positives. In any case, their genomic locations suggest that they would be unlikely to play a role in a conserved artemisinin resistance phenotype because four of these candidates were intergenic (chr01, 03, 07, 10) and the other five (chr07, 12 and bin (three)) were in genes without orthologues in P. falciparum, in contrast to the eight confirmed point mutations.
The two remaining low confidence point mutations mapped to chr05 or contig11844 and had higher quality scores (33 and 31, respectively). However, they were located close to, or within extended regions of low coverage proposed to be deletions (one of which, on chr05, was previously noted) [26], see below. False positive SNPs such as these were previously termed 'proxies for deletions' and their artefactual appearance explained [25].
It is concluded that the I568T mutation in the gene PCHAS_143590 encoding the mu (μ) chain of the AP2 adaptor protein complex is likely to be the only point mutation in AS-ART relative to its immediate precursor AS-30CQ.

Small indels, larger deletions and CNV detection
Forty-one potential insertions or deletions (indels) (14 larger (>3bp) indels, 27 small (≤3 bp) indels) and seven potential copy number variants (CNVs) were called (see Additional files 4, Additional file 5 and Additional file 6) in AS-ART relative to AS-sens. The specific detection and identification of large insertions was not feasible because the single-end sequencing data used in these studies were not able to support the use of more sophisticated detection algorithms that depend upon the availability of paired-end data.
The 27 small indels were considered to be low confidence predictions, ie, false positives. Dideoxy-sequencing confirmed that the five tested were all false positives (see Additional file 4), as in previous studies of AS-15MF and AS-30CQ [25,28]. It is suggested that the majority of these predicted low confidence mutations are therefore false positives. In any case, only six of the other 22 small indels were intragenic.
For larger indels, analysis based on read depth reveals large deletions (say >20 bp) with greater reliability than for intermediate deletions (say 3<bp<20) and all larger insertions. Four (see Additional file 5, highlighted in yellow) of the 14 intermediate and large indels candidates were classified as high confidence mutations (Table 1, Methods), principally because they were predicted using both MAQ and SSAHA2 but also because they generate consistently low read coverage over an extended region. Also, three (chr04, 05, 07) were previously identified in the progenitor clone AS-30CQ or in a related clone AS-15MF [25,26]. One of these was previously confirmed by dideoxysequencing as an intergenic 34 bp deletion on chr07. Another was a potential large deletion (>1 kb) on chromosome 5 located within a gene coding for a non-syntenic (ie, its chromosomal location is not conserved across malaria species relative to surrounding genes) S-antigen (PCHAS_051910-20). The third was a deletion in noncoding sequence at the right hand end of chromosome 4, in a region previously defined to be the site of deletion and translocated copy of a large duplicated chr12 fragment containing mdr1 in the related clone AS-15MF [27]. The remaining high confidence large deletion corresponds to a potential~23 kb deletion on an unassigned contig ('bin' contig11844), not predicted in the progenitor clone AS-30CQ, spanning several non-syntenic genes (PCHAS_000700 to PCHAS_000760). The regions corresponding to deletions on chr05 and in contig 11844 also contain an example of a SNP 'proxy for deletion' (see section above). Verification of the three deletions (in chr04, chr05 and contig 11844) by di-deoxy sequencing was not possible due either to their size or position or both. For the other 10 larger deletion candidates, six are intergenic and four are predicted to interrupt P. chabaudi-specific genes, three mapping to the extreme ends of chr13 (two) and chr14, and one to an unassigned contig (see Additional file 5).
It is worth noting that the frequency of indels in the chicken and human genomes (e.g. [46,47]) is less than the SNP frequency; confirmed in next generation shortsequencing analysis [48]. If applicable to Plasmodium spp, the number of indels will be less than the number of point mutations. Since the number of point mutations in AS-ART (relative to AS-sens) is likely to be eight, the four high-confidence deletions detailed above may represent the full complement of indels. Only one (in contig 11844) appears in AS-ART, relative to AS-30CQ.
For the seven potential CNVs, identified by higher read comparative coverage (range 1.5 -3.1), they extend over very small regions (205-282 bp) and lie outwith coding regions of genes (see Additional file 6). In other clones of the P. chabaudi AS lineage, previously validated largefragment CNVs (involving mdr1) were identified using a similar comparative read-coverage analysis ( [27] and data not shown). In the present case, the seven potential CNVs are unlikely to represent gene duplication/amplification events, simply reflecting natural variation of read coverage.
It is likely that there is only one high-confidence deletion (contig 11844) arising in AS-ART relative to AS-30CQ. The likelihood of other indels or CNVs arising in AS-ART is considered to be low and, in any case, unlikely to confer artemisinin resistance phenotype 2.

Conservation and structure of AP2 μ-chain
An alignment of the predicted amino acid sequences of AP2-μ of P. chabaudi (PCHAS_143590, 597 aa) and its P. falciparum orthologue, PF3D7_1218300 ( Figure 3A) and other Plasmodium spp (data not shown) confirm that this gene is highly conserved in Plasmodium spp. Specifically, the I568T mutation appears close to the C-terminus in a particularly highly conserved region. BLAST searches identify AP2 adaptor μ-chains from other species such as R. norvegicus (435 aa) ( Figure 3B) showing high sequence conservation in both the N-terminal domain (aa 1-154, numbering P. chabaudi unless otherwise stated) and the Cterminal domain (aa 244-597) which forms two subdomains each composed of a β-sheet. These large regions of β-sheets bind to, and hence select cargo protein during coated pit formation and vesicle formation during endocytosis [45,49].
In addition to the conservation of its amino acid sequence, other sequence and structural features of PCHAS_143590 support its designation as the μ-chain of the AP2 adaptor complex and a role in recognition of cargo protein and endocytosis. Residues 155 to 243 comprise Plasmodium specific sequence and extend the linker (rat136 -158) between the AP2-μ N-terminal and C-terminal domains. At its C-terminal end, this linker contains T245, homologous to rat T156 that requires phosphorylation for endocytosis in vitro and in vivo [50]. Also, the P. chabaudi protein contains 17 lysine residues (aa 410 -537) in a region where positively charged patches in rat μ chain interact with phosphatidylinositol bisphosphate (PIP 2 ) [51]. Both phosphorylation and PIP 2 or PIP 3 binding stabilize a large conformational change that allows cargo signal motifs to access the μ-chain signal recognition site at the membrane surface [52].
BLAST alignments ( Figure 3A, 3B) support the designation of P. falciparum I592 and R. norvegicus V401 as the corresponding residues to P. chabaudi I568, suggesting that these residues will have similar positions within the protein structure, and therefore, a corresponding functional significance.
X-ray structures for both the μ-chain alone or the adaptor complex cocrystallised with peptides containing YXXΦ (where Φ represents a hydrophobic residue such as F, I, V, L or M) recognition motifs on the target (cargo) ligands [53] include a rat μ-chain structure (pdb 2PR9), showing that the V401 residue of the rat μ-chain is adjacent to the L9 Φ residue (L, lysine) of the bound DEEYGYECL peptide and, along with rat L173, L175, L404 and V422, forms a hydrophobic pocket into which the Φ residue binds ( Figure 4A). The high sequence conservation between PCHAS_143590 and the rat AP2 μ-chain ( Figure 4B, Figure 3B) suggests that I568 will play a similar role in the P. chabaudi AP2 adaptor to that of V401 in the rat homologue. Homology models of the P. chabaudi structure were predicted by threading the P. chabaudi amino acid sequence into the rat crystal structure, using I-TASSER. The resulting structures indeed show that I568 (P. chabaudi, Figure 4B) and I592 (P. falciparum, Figure 4C) are predicted to lie in a corresponding position, relative to the Φ residue in the bound peptide. The likely effect of the I568T mutation was investigated in the P. chabaudi homology model. The most likely conformation of 568T was predicted by minimising the contact of the side chain oxygen atom in 568T to nearby hydrophobic residues. This results in the oxygen pointing towards the solvent, hence forming part of the Φ-binding pocket ( Figure 4D). The effect of the 568T hydroxyl group is therefore to add some hydrophilic character to the hydrophobic pocket into which the Φ residue fits. Similar results were obtained for the P. falciparum μ2 homology model (data not shown). The I568T mutation is predicted to moderate but not abrogate the binding of the μ subunit to the YXXΦ recognition motif present in the cargo protein. In any case, at least in yeast, an AP-2 μ-chain knock-out is not lethal [54] and clathrin may function independently of the adaptor [55].

Polymorphisms in Plasmodium falciparum ap2-μ gene
Because the only confirmed point mutation arising in P. chabaudi AS-ART along with ART resistance phenotype 2 lies in the AP2-μ gene, the polymorphisms in the P. falciparum orthologue (PF3D7_1218300) were investigated. The DNA sequence analysis of PF3D7_1218300 in 24 samples (see Additional file 2) from Brazil, São Tomé and Rwanda revealed a total of 12 independent polymorphic sites, consisting of nine SNPs and three indels. Seven SNPs and all the insertions were novel genetic variants (relative to existing information in PlasmoDB, Table 2) and unrelated to each other. A polymorphism similar to P. chabaudi I568T was not detected in any of the isolates inspected.
Five of the nine SNPs were non-synonymous (aa 146, 160, 199, 233 and 437) while four were synonymous ( Table 2). The non-synonymous mutations R146K, S160N and F437L occur within regions which appear to be well conserved in Plasmodium spp of the N-terminal domain of AP2-μ ( Figure 3A), although residues in these regions are not well-conserved relative to the rat homologue ( Figure 3A, B). The non-synonymous SNPs, K199T and N233K are situated in unconserved regions. The three indels were three-nucleotide insertions of an AAT codon (asparagine) found adjacent to codon 233, 324 or 329 in asparagine-rich sequence ( Table 2) not conserved in the P. chabaudi orthologue. Amino acids 324 and 329 are located on a P. falciparum-specific region relative to both P. chabaudi and the rat homologue ( Figure 3A, B).

Increased artemisinin resistance phenotype
An ART resistance phenotype (phenotype 1) in P. chabaudi was previously described and the underlying genetic mechanism (mutations in ubp1) identified [25]. That phenotype had been experimentally evolved in AS-30CQ under chloroquine selection and without any exposure to ART. Subsequently AS-ART was evolved from AS-30CQ by ART selection [29]. Here, an increased artemisinin resistance (phenotype 2) in AS-ART relative to AS-30CQ is characterized. The increased resistance is apparent at 200 mg ART kg -1 d -1 (inoculation of 10 6 parasites, three-day drug treatment). Interestingly a five-day treatment using a higher inoculum (10 7 parasites) also differentiated AS-ART from AS-30CQ. These data are consistent with the hypothesis that quiescence or dormancy responses may underlie survival of artemisinin treatment [56,57].

Genetic basis of increased artemisinin resistance
Illumina whole-genome sequencing reveals a single point mutation arising between AS-30CQ and AS-ART. This (See figure on previous page.) Figure 3 Conservation of AP2 μ-chain amino acid sequence. Conservation between P. chabaudi (PCHAS_143590) sequence and A) P. falciparum (PF3D7_1218300) or (B) rat (P84092 UniProtKB/Swiss-Prot) sequences. The P. chabaudi mutated residue I568 and the corresponding P. falciparum residue I592 (A) and Rat V401 (B) are highlighted (purple). Note Plasmodium spp.-specific sequence (~90 residues, relative to rat) lying between positions 154-244 (P. chabaudi). Figure 4 I568T mutation in AP2 μ-chain interacts with YXXΦ motif on cargo protein. A. Rat (P84092) V401 is homologous to P. chabaudi AP2 μ-chain I568, and contributes to the hydrophobic pocket that binds L9 residue of bound peptide (DEEYGYECL) in structure 2PR9. B. Homology model for P. chabaudi wild-type homologue shows similar structure with I568 corresponding to rat V401. C. Homology model for P. falciparum orthologue shows similar structure, with I592 corresponding to rat V401 and P. chabaudi I568. D. Homology model for P. chabaudi mutant -568T is predicted to reduce the hydrophobic character of the binding pocket in which peptide L9 binds. Note increased polar character associated with threonine hydroxyl group in D relative to B. mutation encodes an I568T substitution in the μ subunit of the AP2 adaptor complex (PCHAS_143590). The number of false negatives due to insufficient coverage is estimated to be less than one/genome, as previously discussed [25,26] and given the minimum threshold for point mutation detection (three reads), the proportion of the genome covered by less than three reads (1.67%, see Additional file 4) and the accurate identification of simulated mutations (random alterations in reference sequence) [25]. The frequency of a false negative arising from incorrect read alignment is minimized because the re-sequenced and reference genomes are largely identical, and because optimal hashing-table algorithms were used. Importantly the likely appearance of only one point mutation is consistent with previous genome re-sequencing studies [26] that reveal a consistent set of a small number of point mutations, most of which confer drug-resistance phenotypes in other drugresistant clones of the P. chabaudi AS lineage (Table 1) [25][26][27][28]. For example, four or five out of seven single nucleotide substitutions between AS-sens and AS-30CQ confer resistance to pyrimethamine, chloroquine and artemisinin phenotype 1 [25,26]. This is consistent with similar studies using other parasites; for example, only four nonsynonymous mutations arose in each of two miltefosineselected Leishmania major genomes [58]. Although the possibility that increased ART-R is conferred by unidentified point mutations in the small fraction of the unassembled genome cannot be excluded, regions of low read coverage, the indel identified in AS-ART, or further unidentified indels, suggest that the mutation I568T in AP2-μ on chr14 is the principal determinant of higher-level artemisinin resistance (ART-R phenotype 2) in AS-ART which also bears the ubp1 mutation [25] but no mdr1 amplification [27]. This hypothesis should be validated by transfection experiments.
Other results are consistent with this hypothesis. A previous study suggested that a locus on chr14 contributes to artemisinin responses in genetic selection experiments using AS-ART [23]. Also, two recent independent reports show evidence of the involvement of adaptor proteins in drug resistance phenotypes in other human parasites. A genome-wide RNAi target sequencing approach revealed Trypanosoma brucei parasites with a knockdown of one of each of the four AP1 adaptor subunit genes (as well as lysosomal proteases, vacuolar protein sorting factors, etc.) of the endosomal system in suramin-treated populations [59]. In a similar experimental paradigm to that described here, an in vitro miltefosine-resistant L. major line bore a D762 mutation in the α-subunit of an adaptor protein and a different mutation in the Leishmania infantum orthologue [58]. The probability that mutations in adaptor subunits confer resistance to three structurally and functionally unrelated drugs (suramin, miltefosine and artemisinin) in three different parasites (Trypanosoma, Leishmania and Plasmodium) suggests that a conserved aspect of drug treatment and response is targeted by these mutations.

Clathrin-mediated endocytosis -insights regarding ART action and resistance
The AP2 adaptor complex is a heterotetramer (α, β, μ, σ) that selects and recruits other factors, including membrane protein (cargo) and intracellular clathrin, mediating clathrin-mediated endocytosis (CME) [45]. CME is involved in the internalization of extracellular molecules and ligands (e g, low density lipoprotein, transferrin, growth factors, antibodies and bacterial toxins), the remodelling of the plasma membrane (removal of variable antigen), and membrane protein trafficking and vesicular sorting through the endosomal system and lysosomes [49]. In P. falciparum, these cellular processes are being studied particularly regarding digestive vacuole genesis, haemoglobin uptake and digestion [60], the role of (poly)phosphorylated phosphatidyl inositol [61], the action of aminoquinolines and artemisinins [6,7,[62][63][64] and the trafficking of a critical determinant of chloroquine resistance (the membrane protein, crt) [65]. The participation of clathrin (as coated vesicles) in endocytosis and its relationship to the formation of cystosomes has not been fully clarified and, in any case, the mechanism would require the production of vesicles with double membranes (parasitophorous vacuole and parasite membranes). Nevertheless, it is possible that some elements of CME have been recruited to the haemoglobin uptake mechanism. The existence of Plasmodiumspecific additional sequence between the N-and C-terminal domains of the AP2-μ chain may reflect non-canonical functional attributes of the AP2 adaptor complex.
The I568T mutation in the AP2 μ-chain locates to a residue forming a hydrophobic pocket into which the Φ residue of the YXXΦ recognition motif on the cargo proteins binds. This suggests that modulation of cargo trafficking may be the key event in this resistance pathway. The I568T mutation in AP2-μ may reduce its binding to membrane cargo and mediates artemisinin resistance phenotype 2 by reducing endocytosis or the recycling or trafficking of one or more membrane proteins. Previous work has shown that fluorescent artemisinin derivatives concentrate in neutral lipid bodies close to the digestive vacuole [5], suggesting that drug may be taken up by endocytosis. Artemisinin interacts with the endocytic and endosomal pathway. For example, it inhibits endocytosis [62] and haemoglobin uptake [7], resulting in the disruption of the DV membrane [6] and some derivatives may increase the accumulation of endocytic vesicles [6]. Also, artemisinin activity is dependent on haemoglobin digestion [7]. A reduction in haemoglobin uptake or digestion may therefore constitute a strategy for surviving exposures to short-lived artemisinins. The I568T mutation might therefore reduce the rate of haemoglobin uptake and hence the activation or action of artemisinin. Indeed, the reduced growth of AS-ART in relation to its sensitive progenitor AS-30CQ in absence of treatment (Figure 2A), indirectly supports this notion. The involvement of adaptor proteins in resistance to suramin, miltefosine and artemisinin in Trypanosoma, Leishmania [66,67] and Plasmodium however, may suggest a rather general resistance mechanism involving, for example, the reduction of drug uptake through endocytosis.
A V2728F mutation in ubp1 was previously identified as the critical determinant of artemisinin resistance phenotype 1 in AS-30CQ [25], predicting that this mutation would decrease de-ubiquitination. The ubiquitination of membrane receptors and proteins may induce clathrinindependent (caveolae) endocytosis [68]. Therefore, the ubp1 and AP2-μ mutations may both modulate endocytosis, possibly changing the balance of endocytosis toward a clathrin-independent pathway. Interestingly, chloroquine is also known to interact with endocytosis, vesicular transport, haemoglobin digestion and actin and DV dynamics [62][63][64]69] and the V2728F ubp1 mutation has been shown to confer high-level CQ-resistance [26]. This artemisininand chloroquine-cross resistance is therefore consistent with the suggestion that endocytic pathways are involved in the mode of action of both of these drugs and the resistance mechanisms in P. chabaudi.
The present study therefore suggests that experimental studies on the molecular interactions between ubiquitination pathways, endocytosis and artemisinin uptake and sensitivity may provide important insights regarding artemisinin action and resistance mechanisms.
Field studies of Plasmodium falciparum 12 genetic polymorphisms (nine SNPs and three insertions) were identified in the P. falciparum orthologue (PF3D7_1218300) in field samples from Rwanda, São Tomé and Brazil. Further studies will therefore be required to evaluate the association between PF3D7_1218300 polymorphisms and artemisinin responses. Association studies in locations in Southeast Asia where ART-R has been reported would be of particular interest. A recent linkage analysis of a P. falciparum genetic cross (HB3 × Dd2) revealed three loci linked to in vitro artemisinin responses; on chr05, chr12 and chr13 [24]. Marker C12M63 was the most highly linked marker on chr12. However, this marker maps over 1 Mbase downstream from PF3D7_1218300 suggesting that a different gene was involved in that study. A genome-wide analysis of genetic diversity in P. falciparum parasites from Asia, with different artemisinin responses mapped a possible genetic determinant to a region of chr13 [21]. However, no SNPs were identified in these genes that could be associated with the in vivo parasite's responses to ART. Recently however, the genotyping of P. falciparum isolates from Asia established associations between four SNPs (on chromosomes 10, 13 and 14) and both recent positive selection and parasite clearance phenotypes after artemisinin treatment [70]. However, the genes and regions where these SNPs lie are not linked to the P. falciparum orthologue of the AP2 μ-chain on chromosome 12, suggesting a different gene to that identified here is involved Marian Thomson, Karim Gharbi and other GenePool staff for assistance with sequencing. Members of the Pathogen Genomics group at the Wellcome Trust Sanger Institute and core sequencing and informatics groups gave valuable access to the P. chabaudi genome sequence, assembly and annotation data. We thank the staff of the following institutions for the P. falciparum samples used in this study: Centro Nacional de Endemias, São Tomé, Democratic Republic of São Tomé and Princípe, Secretaria de Vigilância em Saúde, Instituto Evandro Chagas, Belém, Pará, Brazil and the National Malaria Control Program, Kigali, Rwanda. Artemisinin was a kind gift from Dafra Pharma. Wellcome Trust will fund all Open Access charges for any work carried out or partially carried out under one of their grants. Article Processing Charges were supported by grant number 082611: Wellcome Trust Centre In Development (Centre for Immunity, Infection and Evolution). Principal PI: Prof. Rick Maizels.