Association of mutations in the Plasmodium falciparum Kelch13 gene (Pf3D7_1343700) with parasite clearance rates after artemisinin-based treatments—a WWARN individual patient data meta-analysis

Artemisinin-based combination therapies (ACTs) have been the WHO recommended treatment for uncomplicated Plasmodium falciparum malaria since 2006. Studies conducted in 2006–2007 reported that P. falciparum parasites in northwest Cambodia had reduced in vivo susceptibility to artemisinins [1, 2]. This was manifest as delayed clearance of parasites from the blood of patients treated with artesunate monotherapy or ACTs [2]. In 2014, parasites selected in vitro for delayed response to artemisinins were found to have a  mutation in the kelch13 gene (pfk13, P. falciparum 3D7_1343700) [3, 4], a locus consistent with genome association studies that had strongly associated a region of chromosome 13 with slow parasite clearance [5, 6]. Subsequent clinical studies in western Cambodia showed that delayed parasite clearance in clinical trials was associated with several non-synonymous mutations in pfk13. In particular, mutations in the distinctive propeller region of the Kelch 13 protein (codons 441–726) were associated with slow parasite clearance and subsequently with reduced artemisinin susceptibility in in vitro studies assessing susceptibility of the ring-stage parasites [3, 7, 8]. Recently, the structure of the propeller region was solved and made public by the Structural Genetics Consortium [9], making it possible to link particular genetic changes with their position in the molecule. A central role of pfk13 propeller mutations in mediating ring-stage resistance was confirmed by demonstration that parasites engineered to contain the mutations showed ring-stage resistance to artemisinin whereas parasites with the wild type allele were sensitive [10].

As pfk13 propeller sequences were determined in P. falciparum from the Greater Mekong Subregion (GMS), other Asian sites [6, 11‐22], and Africa, Latin America, and Oceania [23‐26]; almost 200 different non-synonymous mutations have been identified (see [27] and supplementary tables 6 and 7 in [28]).

The parasite clearance half-life (PC_1/2), which measures the slope of the log-linear component of the parasite clearance curve, has become established as the best in vivo metric of P. falciparum artemisinin susceptibility [29‐31]. Although sporadic parasite isolates with pfk13 propeller mutations and occasional patients with slow parasite clearance have been identified in many parts of the malaria-endemic world, pfk13 mutant parasites isolated from patients with the slow-clearing in vivo phenotype have been demonstrated only in a circumscribed region of Southeast Asia, initially in western Cambodia and currently extending to parts of Thailand, Vietnam, Lao PDR, Myanmar, and Yunnan Province, China. Within this region, there has been clear evidence of selection and spread of successful artemisinin-resistant parasite lineages [32]. As expected, artemisinin resistance has led to further selection of ACT partner drug resistance [33‐35]. Outside the GMS, extensive molecular analyses have shown no evidence that parasites carrying pfk13 propeller SNPs are under directional selection by artemisinins [23, 25, 27, 28, 36]. Taken together, available studies generally support the conclusion that certain pfk13 propeller mutations mediate delayed parasite clearance following treatment with artemisinin derivatives in Southeast Asia, but that the pfk13 propeller region mutants observed outside the GMS do not show an association with slow parasite clearance [37, 38]. Exceptions include recent molecular reports of independent emergence of parasites that carry pfk13 propeller mutant alleles observed previously in the GMS in Guyana (C580Y) [39] and Rwanda (P574L and A675V) [40]. Accurate measurement of the parasite clearance phenotype requires frequent timed blood sampling [41]. This is often impractical and means that many surveys only collect information on the pfk13 genotypes in a study area. It is therefore critical to utilize optimally those studies that do include both genotype and in vivo phenotype. The WHO has provided guidelines for associating P. falciparum parasite mutations with slow parasite clearance in vivo [42]. Currently, 6 pfk13 propeller region mutant alleles have been validated in this way, 8 additional alleles have been described as “associated” with the phenotype, and 18 other alleles have been isolated in such low numbers that their phenotype could not be evaluated. It is also clear that some non-synonymous mutations in the pfk13 propeller region (notably A578S) are not associated with artemisinin resistance. The low prevalence of pfk13 propeller alleles outside the Greater Mekong Subregion has limited our ability to confirm the roles of additional genotypes in conferring the delayed clearance phenotype.

We have used meta-analysis of individual patient data from published and unpublished studies to compare standardized parasite clearance half-life estimates and pfk13 propeller region genotypes. This analysis capitalized on the power of the large data set to allow assessment of the importance of a broader range of propeller region alleles in mediating slow parasite clearance in vivo.

This is the largest evaluation of the association of different pfk13 mutations with slow parasite clearance following antimalarial treatment with the artemisinin derivatives. Our study confirms that many, but not all (notably A578S), mutations in the propeller region of the pfk13 gene are associated with slow parasite clearance. It also shows that a mutation outside the propeller region, E252Q which had been proposed to confer artemisinin resistance, is strongly associated with slow parasite clearance. This is also consistent with the increased prevalence for several years on the northwest Thailand-Myanmar border of parasites that carried the E252Q mutation, presumably as a result of selection, before the ascendance of the C580Y genotype conferring a more extreme phenotype [18].

Several factors affect the rates of parasite clearance following administration of artemisinin derivatives and would have contributed to the inter-individual and between-site differences observed in this study. In infections with pfk13 wild type parasites, treatment with an artemisinin derivative results in accelerated ring-form clearance, and this is reflected in a steep slope of the parasitemia-time profile and a derived PC_1/2 which is usually well below 5.5 h. Background immunity, or its surrogate, age has a significant additional effect, further accelerating parasite clearance. Thus, immunity as reflected by antibody concentrations can have a relatively small but significant effect [58]. Another important contributor to the half-life is the stage of parasite development at the initial presentation of the patient. This variable, too, could disproportionately affect clearance of pfk13 mutant parasites compared to wild type infections [59] but should only affect inter-individual and not inter-site differences. In artemisinin combination treatments, the partner drug also makes a small contribution to the rate of parasite clearance, which in this study, was significant only for the mefloquine combination, confirming previous findings [60]. It is important to note that despite modest site-specific differences, the clearance half-life derived for each pf13 propeller alleles in the pooled data set was at least 1.5-fold higher than the half-life of the wild type parasites, and for 14 of the 20 propeller mutant alleles, the ratio of mutant to wild type PC_1/2 was between 2.0 and 2.7.

The prominent role of pfk13 propeller mutants in the selective response to artemisinin drug pressure has been recently demonstrated by the rapid increase in parasites that carry a genotype with a particular version of the C580Y allele. These parasites originated in northwestern Cambodian and Thai foci and spread recently to new areas of Cambodia and Vietnam [22, 32]. On a more local scale, in this study, we found that the median PC_1/2 of parasites that already carried a C580Y allele of pfk13 increased from 5.4 h in 2009 to 7.2 h in 2014 in the western border region of Thailand. More widely, among the 14 pfk13 propeller mutants newly associated with slow clearance, 4 of the 14 mutant alleles associated with slow clearance were present in at least 8% of the isolates identified in the corresponding parasite population (range 8–31%). Ten of the newly identified pfk13 mutant alleles were present at low prevalence (range 1–6%) in the sites where they have been reported and even among the validated alleles, prevalence was below 10% for four of the six alleles. Because the pooled data set contains isolates collected from many different sites and over a time span of 14 years, no trends in overall prevalence can be inferred in this analysis. However, the value of following temporal changes in the prevalence of propeller mutants in local sites is clear, as has been demonstrated for parasites that carry the C580Y allele in Thailand, Cambodia, and Vietnam [22, 32].

Detailed temporal studies pfk13 mutant allele prevalence both in the GMS and in other worldwide sites represent a powerful tool for assessing whether a particular pfk13 mutant population is increasing, a signal that further studies of the parasite phenotype may be warranted in that location. Temporal studies will be particularly important for the seven newly identified propeller mutant alleles in our data set that were observed at prevalences of 4% or lower.

In contrast, the pfk13 mutant parasites in all five African sites remained at very low prevalence, generally below 3%, and no evidence of slow-clearing parasites or selection for mutant parasites has been identified. The low prevalence of pfk13 mutant parasites in the African studies in our data set depends on relatively small numbers of parasites (e.g., 31, 26, 29). However, the generality of this observation is supported by extensive studies in many sites in Africa including some assessments of parasite clearance in vivo [37, 38, 61‐63], protection against parasite exposure to artemisinins of cultured parasites in vitro [10, 62, 64], and widespread molecular surveillance of pfk13 propeller region sequence ([23‐26, 55, 65‐68] and see the WWARN K13 surveyor and WHO drug resistance threats map for details). These many reports also support the conclusions of our study: in African sites, pfk13 propeller mutants are diverse but rare, with no evidence for selection even where artemisinin-based antimalarials have been intensively used. Moreover, the pfk13 propeller mutant alleles commonly observed in Asia have rarely been observed in Africa [40, 69]. In addition, only two African sites report a somewhat higher prevalence of parasites with pfk13 propeller mutations, 14% (18/130) in Mali [23] and an increase from 3% (1/31) in 2007 to 27% (7/26) in 2014 in Grande Comoros [70] and 17% (5/29) mutants in individuals from the neighboring island of Mayotte [71].

The pfk13 gene has been identified as essential to P. falciparum in its blood stream stage [72], and it is likely that parasites dependent upon a mutant K13 protein would incur a considerable fitness disadvantage [64]. In this situation, parasite populations subjected to intensive use of artemisinin-based antimalarials would be expected to be under intense selection for pfk13 mutant parasites that carry additional genetic changes that could compensate for the fitness cost as has been observed in other parasite populations under drug selection [73]. There is increasing evidence that such adaptive responses have evolved in Southeast Asian P. falciparum populations [74]. For example, intensive use of the ACT dihydroartemisinin-piperaquine in Cambodia and Thailand has apparently exerted further selection on the parasites that carry the pfk13 C580Y “spreading” genotype, selecting parasites that also carry an increased copy number of plasmepsin 2–3 that confers resistance to the piperaquine partner drug [33‐35]. In addition to selection and spread of pfk13 mutant parasites, independent populations that carry new versions of the common C580Y or one of many novel pfk13 alleles have emerged in many other sites within the GMS region [12‐15, 21, 22, 28, 54].

Even before the pfk13 gene was identified as a potential molecular marker of reduced artemisinin response, genome-wide association studies identified several regions of the P. falciparum genome associated with slow parasite clearance on chromosomes 10 and 14, in addition to chromosome 13 where the pfk13 gene is located [5, 6, 75]. Whole genome analyses have also identified genetic changes associated both with pfk13 propeller mutants and with slow clearance in independent, low diversity populations in Cambodia [76, 77] and in Western Thailand [78]. Parasites that carry these additional genetic changes are present in many populations in the western regions of Southeast Asia [21, 28, 79]. Moreover, the importance of other genetic changes for the expression of the slow clearance phenotype has been observed in vitro. Introduction of a pfk13 mutant allele into a cloned, fast-clearing wild type parasite of recent Cambodian origin conferred more protection against dihydroartemisinin in vitro than when the same allele was expressed in a wild type parasite line of African origin [10]. These observations support the interpretation that the independent emergence of parasite populations with new pfk13 alleles is facilitated in Southeast Asia by the presence of local parasite populations with genomes already adapted to support pk13 mutant parasites when they arise. Exposure to artemisinins could then quite rapidly select emergence and selection of a local population of pfk13 mutant parasites.

These specific adaptive genomic changes that are common in Southeast Asian parasites have not been observed in African isolates, but comprehensive analyses have not yet been undertaken [28]. The slow clearance phenotype is extremely complex, dependent on changes in the overall stress response to artemisinins (see [80, 81] for comprehensive reviews) and there may be many other combinations of adaptive genomic changes that could support the fitness of pfk13 mutant parasites. In addition, it is clear that genetic changes other than pfk13 mutations can confer diminished response to artemisinins in vitro and in vivo [82‐85] and surveillance plans will need to remain open to the possibility of these alternative genetic strategies.

This meta-analysis still depends on detailed multiple quantitative assessments of the parasite count, not an approach that is likely to be feasible for routine surveillance in many study sites. This study highlights the limitations in sensitivity and specificity of the standard assessment of parasite clearance on day 3 as a phenotypic signal of delayed clearance. There is no question that simpler methods for assessment of parasite responses to artemisinins will be needed in all endemic areas, with particular attention to areas where malaria transmission is low and drug pressure will be most likely to select resistance.

Our large data set allowed identification of significant association of pfk13 parasite genotype and PC_1/2 phenotype even when the sample size of parasites with a particular allele was very small. For example, four different mutant alleles were observed in fewer than eight patient isolates. Even in these cases, the analysis allowed a strong genotype-phenotype association to be identified. Among these rare mutant alleles, the ratio of mutant to wild type PC_1/2 ranged from 1.9 to 2.7, ten of the mutant alleles are present in more than 10% of the parasites in the tested isolates, and all but four were observed in more than one study site. This robust outcome demonstrates that analyses of the PC_1/2 in patients can allow identification of mutant pfk13 alleles associated with slow-clearing parasites even when malaria transmission is low and few patients can be assessed. This suggests that some researchers in areas outside the GMS may be able to assess the PC_1/2 phenotype in vivo [11, 37, 38, 61] or parasite responses to artemisinins in vitro [10, 62, 85‐87] as a screen for artemisinin response as some groups have already done.

Finally, 18 data sets were included in this meta-analysis of individual patient data sets, including 4 trials that had not yet been published. We believe that this demonstrates that the engagement of research groups to share individual patient data can provide a unique opportunity for advancement of public health. This approach can combine and facilitate collaborative analysis of critical data sets, defining outcomes that cannot be achieved by only extracting outcomes from multiple single publications. Such meta-analyses can provide evidence for policy makers, while not precluding subsequent individual study publications.

There are limitations of this meta-analysis. Notably, the uneven representation of isolates derived from different sites of origin between Africa and Asia and within Southeast Asia and the imbalance of longitudinal data among the sites. The extensive long-term record of clinical outcomes of artemisinin treatment from patients on the Thai Western Border contributed many more isolates to the available data set than any other site (1291/3087, 42%). However, this rich sample set is itself diverse, contributing information on 31 of the 44 mutant alleles observed in the Asian data set. Moreover, the patients studied were not from a very limited area, but came from sites situated over an extended region of the northwest Thai/eastern Myanmar border. The diversity of propeller mutant genotypes supports the assumption that this large number of isolates from the Thai Western Border is representative of the Eastern part of the whole region that includes Thailand, Cambodia, Lao PDR, and Vietnam.

The slow clearance phenotype that has evolved in P. falciparum parasite populations in response to artemisinin challenge is complex and can involve at least one non-synonymous change in the P. falciparum-specific region of the pfk13 gene, but most frequently, specific single mutations in the propeller region. These different propeller mutations may also compromise parasite fitness, so genomic studies that further define the other genomic changes in parasite populations that support pfk13 mutants to overcome fitness deficits will be needed.

Despite this complexity of the phenotype, the appearance and selection of mutations in the pfk13 propeller region are valuable markers for surveillance of diminished artemisinin responsiveness in parasite populations. This meta-analysis demonstrates that 14 pfk13 propeller mutants, in addition to the 6 previously validated, are associated with prolonged parasite clearance and could be considered for implementation as molecular markers in the Greater Mekong Subregion, as a potential early signal of slow parasite clearance.

The confirmation of the apparent absence outside of Southeast Asia of common pfk13 alleles associated with slow clearance is welcome. However, testing of a more practical and sensitive protocol than persistence of parasites on day 3 that can be a first signal of prolonged parasite clearance is urgently needed. Such an approach should also include vigilance for diminished artemisinin responses signaled by increasing prevalence of parasites with mutant alleles of pfk13 or of other genetic changes that may also diminish parasite responses to artemisinins.