Health impacts of malaria parasite resistance to atovaquone/proguanil treatment: a review and combined analysis

Infection with Plasmodium spp. is a major cause of mortality worldwide, causing 235 000–639 000 deaths in 2015 and 148 000 000–304 000 000 clinical cases of malaria. Most cases are in endemic countries, although malaria is also one of the most frequent causes of morbidity in travellers returning to non-endemic countries. Atovaquone/proguanil (Malarone^®) is a fixed-dose combination often used as a first-line treatment for uncomplicated Plasmodium falciparum infections in non-endemic countries.1^,2 It has been used on a large scale as a treatment in areas where treatment failures of artemisinin combination therapies (TFACT)3 are problematic.4 It is now considered a first-line prophylaxis against malaria for travellers5 and particularly military personnel whose experience of adverse events with mefloquine prophylaxis is becoming increasingly recognized.6 Atovaquone/proguanil is also being studied in a new chemo-vaccination strategy where individuals are exposed to P. falciparum sporozoites and then take atovaquone/proguanil to treat pre-symptomatic infections and generate antimalarial immunity (P. G. Kremsner, unpublished). Taken together with the recent expiry of patent protection for Malarone^®, usage of atovaquone/proguanil is likely to rise in the future.

Atovaquone is a hydroxynaphthoquinone that selectively inhibits the mitochondrial electron transport chain at the cytochrome bc₁ complex of malaria parasites (Figure 1).7 This mechanism of antiparasitic activity is complemented by the individual actions of proguanil and its metabolite, cycloguanil (Figure 1). Proguanil itself has no direct effects on the parasite, but it enhances atovaquone’s ability to collapse the membrane potential of malaria parasites by sensitizing mitochondria to atovaquone.8 Proguanil is converted into cycloguanil by the hepatic CYP2C19 system and cycloguanil inhibits parasite dihydrofolate reductase (DHFR), which is essential for folate production and parasite replication.9

Several mechanisms can potentially influence the efficacy of atovaquone/proguanil for treatment. Mutations in P. falciparum cytochrome b (PfCYTb) (in particular leading to Y268S/C/N) cause atovaquone resistance both in vitro and in vivo.10–12 Interestingly, a recent report, using a rodent model of malaria infection, describes that mutations in Plasmodium berghei CYTb are lethal during transmission of the parasite in the mosquito vector.13 This suggests that these mutations may not be able to spread within a population, although this hypothesis has yet to be demonstrated for P. falciparum in the field. Cycloguanil resistance in parasites is conferred by multiple mutations in DHFR. Polymorphisms in host CYP2C19 also affect proguanil metabolism and can lower cycloguanil concentrations.14

Reports of frequencies of treatment failure associated with atovaquone/proguanil vary, although the risk of failure has not been systematically examined particularly with respect to mutations at codon 268 of PfCYTb. In this systematic review, we examine all original in vivo data where atovaquone/proguanil was used exclusively to treat malaria and relate findings on risk of recrudescence to mutations in PfCYTb and available results from in vitro assays. We also estimate clinical efficacy of atovaquone/proguanil treatment of uncomplicated malaria. Results may impact on existing guidelines for the treatment of uncomplicated malaria.

A total of 282 records were returned using PubMed and 966 using ScienceDirect (Figure 2). The 1248 records were assessed for eligibility, using title, or title and abstract, and 1144 records were excluded at this point, as they did not meet the inclusion criteria. Of the remaining 104 records, 15 duplicate records were excluded. Full text was obtained for the remaining 89 records and assessed for eligibility. Of these, 33 were excluded as they did not meet the inclusion criteria. Thus, 56 articles met the inclusion criteria for this systematic review; within these, 20 included case reports, 29 included group studies and 15 included in vitro/ex vivo data. The case reports and group studies were included in the meta-analysis.

The 29 group studies (Table 1) consisted of 27 with eligible data for atovaquone/proguanil treatment of P. falciparum infection and single studies with eligible data for atovaquone/proguanil treatment of Plasmodium vivax infection and Plasmodium ovale spp. and Plasmodium malariae infection. Together, the 27 P. falciparum studies began with 1960 patients, of whom 1695 were treated and followed up to 28 days (86.5%). A total of 1640 patients were successfully treated up to 28 days, 83.7% of the 1960 original patients and 96.8% of the 1695 treated and followed-up patients. The one P. vivax study began with 25 patients, of whom 19 were treated and followed up to 28 days (76%). Five patients were successfully treated up to 28 days, 20% of the original 25 patients, and 26.3% of the treated and followed up patients. The one study of P. ovale spp. and P. malariae began with six patients and all were successfully treated up to 28 days.

Of note, only 14 of the studies were RCT designed to test the efficacy of atovaquone/proguanil or used atovaquone/proguanil as a control treatment and participants of these made up only 55% of the total participants included here. Most of the studies from which these data were gathered, including the RCT, were of low methodological quality, being small and having between 18 and 253 participants receiving atovaquone/proguanil. Risk of bias during selection was determined to be unclear in 10 of 14 RCT group studies, as methods for randomization and concealment of allocation were unclear (Table 2). Risk of bias during performance was determined to be high in 13 of 14 studies, as blinding of participants and researchers was used in only one study. Risk of detection bias was determined to be unclear in all but one RCT study, as allocated interventions were not blinded. Risk of bias due to a high rate of attrition (<10%, low; between 10% and 20%, medium; >20% high) or patients withdrawn from the trial without explanation was high in only one RCT study. Risk of bias due to selective reporting was low to medium in all studies as 28 day cure rate was defined as either a primary (low) or secondary (medium) outcome in all cases. Another potential bias was that 11 of the 14 RCT studies were carried out by, funded by or supported by GlaxoSmithKline or its preceding companies Glaxo Wellcome and Wellcome Research Laboratories.

High-quality data for the efficacy of atovaquone/proguanil are scarce, but provide estimates of treatment success in RCT group studies of between 89% and 98% for P. falciparum malaria (Table 1; weighted averages based on population size and heterogeneity), between 20% and 26.3% for P. vivax malaria (from one study) and 100% (in three patients each) for P. malariae and P. ovale spp. malaria.

Comparator antimalarial treatments (with number of times trialled in parentheses) were chloroquine (two), amodiaquine (two), sulfadoxine/pyrimethamine (three), chloroquine/sulfadoxine/pyrimethamine (one), quinine (one), quinine/tetracycline (one), halofantrine (two), mefloquine (one), and the artemisinin-based combination therapies (ACT), artemether/lumefantrine (two), artesunate/mefloquine (one), artesunate/amodiaquine (one) and dihydroartemisinin/piperaquine/trimethoprim/primaquine (one). Nine of the 14 RCT presented here were analysed in a previous Cochrane Library systematic review from 2005.16 Subsequent RCT involving atovaquone/proguanil have used ACT predominantly as the comparator treatment(s). Given the diversity of treatments used in the trials and to allow results to be generalized to a larger population, trial data involving ACT, 4-aminoquinolines (chloroquine and amodiaquine) and amino alcohols (mefloquine, halofantrine and quinine), were grouped for a meta-analysis (Table S1). Sulfadoxine/pyrimethamine was analysed alone. The analysis indicates that there is no significant difference (P =0.83) in treatment success between the use of atovaquone/proguanil and ACT (Figure 3a). Sensitivity analysis was consistent with this outcome (Table S2). Given the grouped ACT in this analysis, we combined the data for two different ACT in one three-arm study.17 However, analysing each arm separately did not change the outcome of the analysis (Table S2). Analysis of atovaquone/proguanil versus the amino alcohols group (Figure 3b) indicates that treatment success with atovaquone/proguanil is not significantly more effective (P =0.18) and statistical significance was maintained for the majority of scenarios during sensitivity analysis (Table S2). As previously reported individually for amodiaquine and chloroquine,16 meta-analysis of the three trials that used atovaquone/proguanil versus 4-aminoquinolines (Figure 3c) suggested that atovaquone/proguanil is more effective than 4-aminoquinolines (P <0.00001) and the sensitivity analysis was predominantly consistent with this outcome (Table S2). This can be explained by the prevalence of mutations in pfcrt and pfmdr1 conferring resistance to chloroquine and amodiaquine in the regions of study.18–20 Similar findings (P =0.001) emerged when analysing atovaquone/proguanil versus sulfadoxine/pyrimethamine (Figure 3d and Table S2). This can be explained by the increasing development of sulfadoxine/pyrimethamine resistance over time between the two studies undertaken in Zambia.21^,39

Eligible data on in vitro/ex vivo clinical isolates exposed to atovaquone were available in 15 papers (Table 3). The amount of data and the level of detail available did not allow further mathematical syntheses, but the data can be used to hypothesize about what the in vitro/ex vivo phenotype of atovaquone resistance might be. All P. falciparum isolates with the WT Y amino acid at codon 268 have an atovaquone IC₅₀ ≤28 nM, with the majority <10 nM. All single isolates with N, C or S at 268 have IC₅₀ values between 20.5 and 17 000 nM. A further four isolates with S at 268 were reported to have a median (IQR) IC₅₀ value of 5.7 nM (1.7–1216).22 Isolates with mixed genotypes were susceptible to atovaquone in vitro, with median IC₅₀ values between 4.7 and 5 nM. Isolates of unknown genotype ranged in IC₅₀ values from low nanomolar to low micromolar. The 38 P. vivax isolates had a pooled mean IC₅₀ value of 29.4 nM.23

Data for case reports were available from 20 papers for 36 individuals (Table 4). Thirty-three of the cases were of P. falciparum infection and there was one case each of P. malariae, P. ovale spp. and P. vivax infection. Variables have been summarized, with means, standard deviations (SD), medians and IQR for continuous or count data and proportions for categorical or binary data types (Table S3). Data for pretreatment parasitaemia (baseline), parasitaemia at treatment failure/recrudescence and genotype were not available for non-falciparum infections and so these species were not included in subsequent analyses.

A raw data plot, Figure 4(a), presents the minimum number of days to recrudescence of infection after atovaquone/proguanil treatment, which takes into account the onset of symptoms if prior to parasitological diagnosis, versus the absence or presence of mutation (Y268S/C/N) in PfCYTb at the time of recrudescence. This suggests that distributions may differ across groups by mutation (confirmed by a preliminary Kruskal–Wallis test; P <0.001). In a subset of parasite isolates it was possible to define if there had been a change in codon 268 following treatment. A raw data plot of the minimum number of days to recrudescence versus this dataset suggested distributions may differ by codon 268 change (P =0.009; Kruskal–Wallis test; Figure 4b).

Figure 5 presents the relationship between pretreatment parasitaemia and minimum days until recrudescence in the absence or presence of a mutation in PfCYTb, using an interaction model (Figure 5a and b). Analyses of the complete and observed (by multiple imputation) datasets suggest that pretreatment parasitaemia does not appear to influence the minimum days until recrudescence in general and that there is evidence that this effect is not modified by the presence of mutation in PfCYTb (P =0.62 and 0.87, respectively; Table S4). However, according to complete data analysis, there is evidence (P <0.001; Table S4) that grouping (the codon 268 present post-treatment) is a statistically significant predictor of the minimum days until recrudescence and the evidence is further supported by the observed data analysis (P =0.002; Table S4). The model predicts that patients presenting with a baseline parasitaemia of 1% will have an average minimum number of days until recrudescence of 29 (95% CI: 22, 35) days if mutation in codon 268 in PfCYTb is present, whilst this is 19 (95% CI: 7.3, 30) days shorter in duration if the mutation is absent. Note that although a slight departure from normality for the standardized residuals (P =0.02) was calculated, we opted for model simplicity rather than introducing another quadratic term.

Figure 5 also presents the relationship between baseline pretreatment parasitaemia and parasitaemia at recrudescence (post-treatment parasitaemia) in the absence or presence of a mutation in PfCYTb, using an interaction model (Figure 5c and d). Analyses of the complete and observed datasets suggest that baseline parasitaemia (on a log scale) increases slightly and linearly with parasitaemia at recrudescence of infection (P =0.004 and 0.029, respectively; Table S5). Furthermore, analysis of the complete dataset suggests that the level of increase differs by grouping using codon 268 presence post-treatment, although this effect no longer holds when observed data analysis has been implemented (P =0.04 versus P =0.217; Table S5). Note that the two settings do not exhibit massive differences in estimates and their precisions. Here, the model predicts that patients presenting with a baseline parasitaemia of 1% (geometric mean, which coincides with the median; Table S5) will have an average post-treatment parasitaemia of 2.0% (95% CI: 1.2%, 2.8%) if a mutation in codon 268 in PfCYTb is present.

Additional analyses to incorporate pretreatment parasitaemia interval values as <0.01 and <5 (Table 4), using scenarios in which these values were ‘1’, their upper limit, ‘2’, half the interval values and ‘3’, a 10th of the value, provided no substantial quantitative changes in the above estimates presented and their precision and no qualitative changes to the conclusion (Table S6 and Table S7).

Atovaquone/proguanil was developed as a combination therapy when early clinical studies showed that atovaquone as a single agent was associated with recrudescence of highly atovaquone-resistant infections in ∼30% of patients.24In vitro evidence of synergy with proguanil prompted development of this combination, whose initial high cost precluded widespread use. As generic formulations of atovaquone/proguanil reduce costs, and as TFACT emerge, atovaquone/proguanil is one of the few non-ACT combinations registered for management of malaria. Determining its overall efficacy and identifying markers that predict treatment failures is important for policymakers in public health.

To carry out the widest scrutiny of evidence on the efficacy of atovaquone/proguanil, we included two broad types of studies. The first type (summarized in Table 1) describes efficacy of atovaquone/proguanil in the treatment of malaria often (in just over 50% of cases) in the context of an RCT. The quality of these types of studies is relatively low for several reasons associated with potentials for bias (Table 2). The second more mechanistic analysis of atovaquone/proguanil’s efficacy (summarized in Tables 3 and 4) included review of in vitro susceptibility analysis of parasites, where available, and detailed analysis of individual case reports of treatment failures and their association with parasitaemia and mutation in PfCYTb. These latter reports are often richer in data and provide insights that complement findings from larger studies.

While datasets were small and associated with potential bias (and thus requiring cautious interpretation), the overall efficacy of atovaquone/proguanil expressed as a weighted average based on study population sizes and heterogeneity is 89% and 83% in ITT analyses of RCT and observational studies, respectively, and is 98% and 99% in PP analyses. This is a reassuringly acceptable level of efficacy and to date there are no indications of treatment failures becoming associated with particular geographical areas that would preclude atovaquone/proguanil use to treat travellers or prevent infections from such areas. Furthermore, meta-analysis suggests that atovaquone/proguanil treatment success is equivalent to the use of ACT and amino alcohols and better than 4-aminoquinolines and sulfadoxine/pyrimethamine, although caution is required in some cases due to the grouping of different antimalarials within a class. This extends findings from a prior meta-analysis that concluded that atovaquone/proguanil is more effective than chloroquine, amodiaquine and mefloquine.16 This general reassurance is important particularly in light of complications that are being associated with the use of mefloquine and that have been reviewed recently in a UK House of Commons Defence Committee report on mefloquine’s use in military personnel.25 Doxycycline and atovaquone/proguanil remain as the only alternatives to mefloquine recommended for antimalarial prophylaxis.5 While atovaquone/proguanil is considered safe, it has been reported that safety data are relatively sparse and would benefit from further large trials.16 The safety of atovaquone/proguanil was not studied here.

The in vitro phenotypic assays for atovaquone susceptibility and its relationship to target genotype suggest that WT amino acid (Y268) is uniformly associated with susceptibility. The threshold for defining susceptibility is an IC₅₀ value ≤28 nM, with most isolates in different studies having IC₅₀ values <10 nM. Although the aggregated IC₅₀ values for P. vivax were 29 nM, it is unlikely that this slightly higher value compared with P. falciparum susceptibility contributed to the higher treatment failure rates as these are most likely due to relapse because of the non-susceptibility of hypnozoite stages found in the liver to atovaquone/proguanil.26

Analysis of individual case reports and the dynamics of recrudescing infection highlight further interesting findings. The presence or appearance of mutation (Y268S/C/N) in PfCYTb is strongly associated with a late recrudescing infection (Figures 4 and 5) where late onset of symptoms or parasitological recrudescence (whichever is earlier, which we have defined as minimum days to recrudescence here) is on average 29 days (95% CI: 22, 35) after treatment has commenced. This is in accord with a previous estimate of the mean time to recrudescence of parasites carrying the Y268C mutation of 28 days (95% CI: 23.0, 33.0).27 Understanding the mechanisms that account for the length of time until recrudescence is worthy of further investigation. One possible factor underlying this phenotype is a loss of parasite fitness due to mutation. This has been reported previously, using in vitro growth assays, for atovaquone-resistant parasites carrying PfCYTb mutations, though not at position 268.28 Our data suggest that patients should be monitored for up to 42 days. Late recrudescence in these cases should always be treated with an alternative antimalarial treatment regimen.

A recent report has demonstrated that mutations in P. berghei CYTb are invariably lethal to the parasite during transmission in the mosquito vector.13 This finding lends weight to the hypothesis that PfCYTb mutations may not be able to spread within a population. If true, this would preclude the requirement to monitor for these mutations in endemic areas. The available data are in general agreement with this, as codon 268 mutations are very rarely observed in parasites from patients that suffer later recrudescence, prior to drug pressure (Table 4) and no geographical foci of atovaquone/proguanil treatment failure or PfCYTb mutations have been reported. However, this does not preclude the spread of PfCYTb mutations carried by parasite sub-populations, where the mutation cannot be detected by conventional means, or the spread of parasites with permissive genetic backgrounds that favour PfCYTb mutation following drug pressure. Our findings also identify the need for further characterization of the genetic backgrounds of parasites in patients experiencing early recrudescence. These studies should aim to determine the mechanism of this high-grade resistance as well as identifying associated markers, although other factors that may cause or contribute to the phenotype of early treatment failure will need to be considered carefully (e.g. non-compliance to treatment, use of substandard or counterfeit medications, poor absorption or metabolism of the medication by the patient).

While not considered in detail, it is worth noting that there are 17 case reports that provide molecular markers for cycloguanil resistance, the triple PfDHFR mutation S108N, N51I, C59R (Table 4). Only 4 of 17 infections carried parasites with sensitive genotypes at first presentation. One of these four infections recrudesced with parasites carrying a resistant genotype, leaving three infections caused by parasites with PfDHFR-inhibitor sensitive genotypes post-treatment. Interestingly, all parasites defined as recrudescing by day 3 (Table 4) carried PfDHFR sensitive genotypes, suggesting that cycloguanil did not contribute to failure. All later treatment failures (from day 7) were caused by parasites carrying genotypes associated with resistance to cycloguanil. Therefore, atovaquone/proguanil treatment failures from day 7 onwards are most likely to be caused by parasites that are already resistant to cycloguanil.

After our database search was closed, an additional series of case reports that was not picked up was identified independently.29 These six cases were of patients who had recrudesced more than once after atovaquone/proguanil treatment and in all cases time to recrudescence was ≥19 days. In five cases where the post-treatment genotype of PfCYTb was available, it was of the 268C/S mutation. In four of six patients with second recrudescences, the time to recrudescence was ≥20 days and all four genotypes bore mutant variants at position 268. These observations suggest that the proguanil component of atovaquone/proguanil has sufficient antimalarial efficacy to suppress parasitaemias for 2–3 weeks and that the dynamics of late treatment failure are consistent with absence of atovaquone efficacy. These cases were incorporated into a secondary analysis of the case reports. Findings with regard to the relationship between pretreatment parasitaemia and minimum days until recrudescence in the absence or presence of a mutation in PfCYTb are consistent with those presented in Table S8.

Overall, atovaquone/proguanil therapy is comparable in efficacy to ACT used in treating uncomplicated malaria. Detailed genotype–phenotype analysis in this systematic review has illustrated several new findings. There are differences between early and late treatment failures because mutations in the target conferring resistance to atovaquone are identified most commonly in late and not early treatment failures. The mechanism of early treatment failure after atovaquone/proguanil treatment needs further investigation. Recent evidence is also reassuring that spread of the 268 mutations conferring atovaquone resistance may be limited by poor transmissibility in the insect stages of P. falciparum infections.