Interactions between antimicrobial agents provide clues as to their mechanisms of action and influence the combinations chosen for therapy of infectious diseases. In the treatment of malaria, combinations of drugs, in many cases acting synergistically, are increasingly important in view of the frequency of resistance to single agents. The study of antimalarial drug interactions is therefore of great significance to both treatment and research. It is therefore worrying that the analysis of drug-interaction data is often inadequate, leading in some cases to dubious conclusions about synergism or antagonism. Furthermore, making mechanistic deductions from drug-interaction data is not straightforward and of the many reported instances of antimalarial synergism or antagonism, few have been fully explained biochemically. This review discusses recent findings on antimalarial drug interactions and some pitfalls in their analysis and interpretation. The conclusions are likely to have relevance to other antimicrobial agents.
Antimicrobial drugs are said to interact when the potency of two or more drugs combined in an assay of antimicrobial activity is either higher (synergism, synergy, potentiation) or lower (antagonism) than would be expected from the individual activities simply added together. Interaction in this sense does not usually, but might, imply actual physical contact between the different drugs in question. More often, the activity of a drug A is affected by drug B and/or vice versa as a result of the molecular interactions of the drugs with components of the target cell and the biochemical/physiological consequences of these interactions. It follows, therefore, that the existence and strength of drug interactions may tell us something about what the individual drugs do to cells. Studies of drug interactions in vitro and in cultured cells are therefore commonly used as an adjunct to mechanistic studies on antimicrobial drugs. More important from the point of view of controlling infectious diseases are the effects of drug interactions on combination therapy (or prophylaxis) with antimicrobial agents. It is common to use two or more antimicrobial agents in combination in order to avoid treatment failure with a single agent. If the combination is synergistic it is particularly favoured because the doses of the drugs may be reduced, maintaining the antimicrobial effect while minimising side effects. Conversely, combination therapy with two drugs known to be antagonistic is generally avoided. It is therefore important to know the nature of the interactions between drugs in clinical use and this information is used in the choice of new drug combinations.
Interactions between antimalarial drugs of various classes have been known for many years. This review will be concerned mainly with the data obtained since antimalarial synergism and antagonism was reviewed by Vennerstrom et al. . To begin with, it is important to discuss some issues in the analysis of data pertaining to antimalarial drug interaction. Often, this analysis is poorly done and statements about synergism or antagonism may not be justified. Although the following section relates only to antimalarial drugs, students of other antimicrobial drug interactions will recognise some of the pitfalls of measuring drug interactions.
2Analysis of antimalarial drug interactions
Antimalarial drug1 interaction may be studied at different levels: in vitro in cell-free extracts or using mixtures of isolated macromolecules, in cultured cells, in animal models of malaria, or in human patients. Of course, the relevance of these levels to antimalarial therapy increases towards the end of the list, but most data are available for drug interactions in cultured parasites and most of these use the asexual, erythrocytic stages of Plasmodium falciparum. Animal models of P. falciparum malaria are confined to certain simian species, so rodent malarial parasites such as P. berghei and P. yoelii are usually the first to be used in drug evaluation . However, care needs to be taken in extrapolating the rodent malaria data to the human malarial parasites.
Human malaria is caused by four Plasmodium species, of which P. falciparum causes the most morbidity and mortality . The complex developmental cycle of the parasite takes it through an anopheline mosquito vector and into the human host via the saliva of the insect. Sporozoites injected during a blood meal invade hepatocytes and after intracellular growth and multiplication new parasites are released as merozoites. Drugs that kill liver-stage parasites are often called tissue schizonticides. Merozoites invade erythrocytes, in which they can follow one of two developmental pathways. The majority undergo growth and asexual multiplication and are released as new merozoites. This cycle, which lasts ∼48 h in P. falciparum, is associated with clinical disease and contains the target stages of most antimalarial drugs. Drugs acting on asexual, intraerythrocytic stages are often called blood schizonticides. The alternative pathway of development in erythrocytes leads (after ∼14 days in P. falciparum) to the formation of male and female gametocytes, which are responsible for sexual reproduction and transmission through mosquitos. Some drugs have activity on gametocytes.
Drug interaction studies using cultured, mainly asexual intraerythrocytic P. falciparum are easy to set up once the culture techniques have been mastered. Generally, drugs are titrated in multi-well plates, parasitised erythrocytes are inoculated into the wells, and a convenient read-out of growth is obtained after incubation for 24–96 h. Various read-outs are available, including incorporation of radiolabelled metabolic precursors such as [3H] hypoxanthine, activities or quantities of marker proteins such as lactate dehydrogenase or parasite numbers and morphologies on Giemsa-stained smears and are reviewed in Noedl et al. . Most of these methods lend themselves to the testing of arrays of concentrations and combinations. Analysis of the data obtained is not so straightforward. Data are most often shown either in a table or bar chart such as the one shown in Fig. 1(A), or as an isobologram (see below). The problem with the former style of presentation is illustrated in Fig. 1(A) and (B), which are based on real data. In Fig. 1(A), it can be seen that the combination of A and B has superior activity to A or B alone. It is commonly concluded in the literature, on the basis of similar data, that A and B are synergistic. However, consideration of the inhibition curve shown in Fig. 1(B) and the fact that in this case, drugs A and B are identical and therefore cannot be synergistic, shows the flaw in this logic. In general the use of this kind of presentation should be avoided, though it may be possible to illustrate antagonism in this way.
Presentation of data in an isobologram (illustrated in Fig. 2) is far better. This type of graph, which dates back over 100 years, has the potential to show not only the difference between synergism (or antagonism) and mere additivity, but also the degree of synergism (or antagonism). Median inhibitory concentration (IC50) values for drug A in the presence of various concentrations of drug B are plotted on the x-axis and vice versa on the y-axis, using linear scales. A line of best fit, which intersects the x-axis at the IC50 for drug A alone and the y-axis at the IC50 for drug B alone, is drawn. (Alternatively, isoboles can be constructed for IC90 or any given effect.) A straight line indicates additivity, a concave line (displaced to the bottom left) synergism, and a convex line (displaced to the top right) antagonism. The strength of synergism (or antagonism) is indicated by the degree of deviation from the line of additivity. This can be expressed in algebraic fashion, most often as a (minimum) fractional inhibitory concentration (FIC). The FIC is defined as the concentration of inhibitor present in the combination divided by the concentration of inhibitor alone that gives the same effect . The sum of FICs of drugs A and B at the mid-point of the curve in Fig. 2 (ΣFIC) is 1 for an additive combination, <1 for a synergistic one, and >1 for an antagonistic one. Sometimes the axis labels of isobolograms are expressed as FIC instead of IC50 but the effect is the same. This approach can be extended to combinations of three or more agents .
For strongly synergistic combinations such as that of sulphadoxine and pyrimethamine (see below), the line is highly concave, the ΣFIC ≪ 1 and there is no doubt of a positive interaction between the two agents. The problem arises when the results are less clear-cut. Lines that are only moderately concave (or convex) leave some doubt as to whether the combination is significantly synergistic (or antagonistic). In response to this difficulty, a ‘cut-off’ value of ΣFIC = 0.5 has been widely used, such that ΣFIC < 0.5 is regarded as synergistic and ΣFIC > 0.5 not. The corresponding cut-off used for antagonism is either 2 or 4. It is important to remember that these values are completely arbitrary. It is absurd to suggest, for example, that a combination that gives a consistent ΣFIC of 0.6 is, on that basis alone, not synergistic. The only meaningful ‘cut-offs’ are those for statistical significance (e.g., at p= 0.05), as in any other type of experiment. This fact has eluded many investigators and editors, who continue to adhere to the ΣFIC 0.5 and 2/4 values as thresholds between interaction and no interaction. This author's opinion is that these ‘cut-offs’ should be avoided and the data subjected to proper statistical analysis. The correct statistical analysis of interaction data presents particular problems that are beyond the scope of this review, but readers are referred to the work of Greco et al.  and Machado and Robinson . The application of Machado's model to antimalarial data is described in a study of protease inhibitors by Gavigan et al. . This ‘response-surface’ model does not use ΣFIC as such but a related quantity η (which also =1 for additivity, <1 for synergism and >1 for antagonism). It has the advantages of analysing the entire set of data points rather than only the derived IC50 and of taking into account experimental error. Other types of analysis have been used in some of the publications cited below but the isobologram is probably the best.
It should be evident from the discussion above that the widest range of concentrations/ratios (i.e., chequerboard design) will give the maximum amount of data for isobologram construction and analysis and is preferable to a limited number of fixed concentrations. Furthermore, interaction may differ at different ratios of drug A:drug B (see example of a skewed isobole in Fig. 2) and experiments employing only limited fixed-ratio combinations may be of limited use. Data from animal models are likely to be more predictive of efficacy in patients and some investigators have found that pharmacokinetic effects lead to different conclusions in vivo from those made with cultured cells. The difficulty of course is that animals are more variable than wells in a culture plate and cannot be used in the same numbers, so chequerboard-type titrations and isobolograms are more labour-intensive to obtain, as should be apparent from the relative sizes of Tables 1 and 2 below. In clinical studies, the experimental design is (understandably) rarely of the type suitable for generating isobolograms or similar analyses. Typically, the efficacy of compound A + compound B is compared with either compound alone (or other treatments) so only cases of strong antagonism (e.g., where the efficacy of A + B is less than that of A alone) would be apparent. A hybrid approach employed by some investigators is to measure interactions using cultured parasites in the presence of those drug concentrations/ratios achievable in humans (or failing that, in animals) according to pharmacokinetic studies.
Table 1. Synergistic and antagonistic interactions between antimalarial agents in cultured P. falciparum
Test strain (nature of testa)
aUnless otherwise stated, inhibition of growth/development of blood-stage parasites was tested.
bA, antagonism; A*, antagonism in the strains listed but not in one or more other strain(s); S, synergism; S*, synergism in the strains listed but not in one or more other strain(s).
Table 2. Synergistic and antagonistic interactions between antimalarial agents in animal models
Test system (nature of testa)
aUnless otherwise stated, efficacy against blood-stage infections in mice was tested.
bA, antagonism; A*, antagonism in the species/strains listed but not in one or more other species/strain(s); S, synergism; S*, synergism in species/strains listed but not in one or more other species/strain(s).
3Relevance of drug interactions to mechanisms of antimalarial action
As stated above, the observation of synergistic or antagonistic interactions between antimicrobial drugs is reflective of relationships between their actions on cellular components. For example, synergism between two antimicrobial drugs might result from (i) binding to the same target protein such that a conformational change caused by the binding of drug A enhances the binding of B, (ii) binding of drug A to a transporter causing increased uptake of B into the cell or the subcellular compartment in which it acts, (iii) formation of a complex between A and B of enhanced toxicity, (iv) stimulation by A of the conversion of B to a more active form, etc. Converse examples could be constructed for antagonism. Therefore, interactions are often measured in the context of mechanistic studies and the presence of synergism or antagonism used as evidence for (or against) a particular mechanistic hypothesis. Therein lies another problem with many studies on antimalarial drug interactions, which is that synergism (or antagonism) can be taken to mean almost anything the investigator wants it to mean. For example, synergism between two agents is often taken as evidence that the relevant targets of the two agents form components of a common pathway. This may be true, but it has been argued that sequential inhibition of linear reactions in metabolic pathways by two or more inhibitors in the steady state alone cannot be synergistic (see discussion of antifolates below), so it is not necessarily so. Another liberty taken is to conclude from data showing antagonism between drugs A and B that the two drugs have a common target. One could also argue (as in example (i) above) that the same conclusion could be drawn from a synergistic interaction, or (returning to the case where A = B) from no interaction. Therefore, in the discussion of specific antimalarial drugs below, mechanistic inferences made from drug-interaction data will be treated with caution unless there is a logical explanation and at least some supporting evidence for the effect.
4Relevance of drug interactions to clinical treatment and prophylaxis of malaria
The use of combinations of antimalarial drugs in the clinic is favoured for various reasons, including the delayed emergence of resistant parasites compared with monotherapy . It is therefore accepted that two or more drugs to be used in combination should be no worse than additive; synergism is preferable. It is important, therefore, to be able to use experimental drug-interaction data to predict likely interactions in the human host. Selection of combination partners (such as proguanil for atovaquone: see below) has often depended in part on observations of synergism in experiments. Such interactions may be predictable from data on cultured parasites, or may emerge only when drugs are used in animals. Questions of statistical significance and in vivo relevance are, therefore, more than purely academic. One point that is seldom mentioned in this context is that synergism does not necessarily widen the therapeutic window in our favour. Two drugs may act synergistically (or antagonistically) on a pharmacodynamic target in the host or may affect one another's pharmacokinetics. When seeking to extrapolate from in vitro or cultured-cell assays to whole organisms, therefore, it is relevant to test drug interactions in the relevant host assays as well. Pharmacokinetic interactions between established antimalarial agents are discussed in Giao and de Vries .
5Interactions between antimalarial drugs
This section deals with well-documented antimalarial drug interactions with possible mechanistic or clinical significance. It takes as its date cut-off point the review by Vennerstrom et al. . The only results included here are those in which the data clearly support the conclusion that there is either synergism or antagonism in at least one strain or infection model. In most papers, there is no satisfactory statistical analysis but the results are included if they are convincing in terms of the degree of deviation from additivity and/or the reproducibility of the observations. Where conclusions about synergism or antagonism are not clearly justified by the data and methodological descriptions provided, they are not included here. There can also be strain differences in the results2 as indicated in the tables and even differences depending on the level of inhibition used for analysis (e.g., IC50 vs. IC90). In the latter case the predominating interaction is taken. The interactions found are shown in Tables 1 and 2 and the relevant mechanistic and clinical implications are discussed below. The list is not necessarily comprehensive.
5.1Interactions among quinoline and related agents
Various synergistic and antagonistic interactions among quinolines (e.g., chloroquine, quinine, mefloquine, amodiaquine) and related agents (e.g., halofantrine, pyronaridine) have been reported (; Tables 1 and 2) and the picture that emerges is quite confusing. One observation that is somewhat consistent is antagonism between chloroquine and mefloquine (and, to a lesser extent, chloroquine and quinine) and the perhaps related antipodal susceptibility to these two agents, i.e., some strains selected for increasing chloroquine resistance become increasingly susceptible to mefloquine and vice versa (reviewed in Foley and Tilley ). Mefloquine can inhibit uptake of chloroquine, giving a plausible basis for the antagonistic effect. Famin and Ginsburg  have shown that the ability of chloroquine to cause accumulation of undigested haemoglobin is antagonised by mefloquine. Host erythrocyte haemoglobin is normally degraded in an acidic, parasite organelle, the digestive (food) vacuole. The haem (ferriprotoporphyrin IX) component represents a toxic waste-disposal problem for the parasite, which is solved by biomineralisation of haem into an insoluble complex, haemozoin. Interference with this process by chloroquine, and possibly other drugs, is believed to lead to membrane damage and death . Famin and Ginsburg hypothesise that mefloquine (and possibly quinine) inhibit(s) endocytosis of erythrocyte cytosol by the parasite, leading to reduced concentrations of free haem, to which chloroquine binds, in the digestive vacuole.
A surprising interaction takes place between amodiaquine and its active metabolite desethylamodiaquine . From ∼1 h after administration, the blood level of the metabolite surpasses that of the parent compound and the ratio of desethylamodiaquine/amodiaquine eventually reaches 500–5000:1. The synergism between the two, in culture at least, is very strong (ΣFIC as low as 0.04) and suggests that amodiaquine may constitute in effect a combination therapy. It is not known whether the two compounds have the same or distinct mechanisms of action but the synergism data show that they are mechanistically not interchangeable.
5.2Interactions among antifolates
Synergistic interactions between combinations of antifolates, generally an inhibitor of dihydropteroate synthase (DHPS) plus an inhibitor of dihydrofolate reductase (DHFR), have been extensively reported (, Table 1). This synergism is so strong and so consistent that these agents are seldom used other than in combination. Examples in current use for malaria include sulphadoxine + pyrimethamine and dapsone + chlorproguanil. Similar examples can be found in other areas of antimicrobial chemotherapy. In spite of the long history of antifolate synergism, there is still a debate as to how it works. The discussion here is confined to Plasmodium but it would be satisfying, though perhaps impossible, to postulate a single mechanism of antifolate synergism that would apply in all susceptible organisms. The widespread assumption that it is the sequential inhibition of two enzymes in the same pathway that confers synergism does not stand up to experimental or theoretical examination . It was therefore suggested by Watkins and co-workers  that synergism between sulphadoxine and pyrimethamine was the result of a second site of action of sulphadoxine on DHFR. This contention was supported by Sirawaraporn and Yuthavong  but their experiments were performed in sulphadoxine concentrations that were probably too high to be relevant . In addition, this suggestion would predict a minor role for dhps mutations in pyrimethamine + sulphadoxine treatment failures, a prediction that has not on the whole been borne out (see  and references therein). A hypothesis more consistent with the available data has been put forward by Hyde . To understand it one must appreciate that most P. falciparum strains can utilise exogenous folates, greatly reducing the requirement for de novo synthesis via DHPS and other enzymes, and in these strains the activity of sulphadoxine is antagonised by relatively low concentrations of folate, close to the normal, physiological average (the ‘folate effect’). In two such strains, 1B-B5 and QC-13, synergism between sulphadoxine and pyrimethamine was very much reduced in the absence of folate or folinate . It was suggested that pyrimethamine has a second site of action related to the uptake or utilisation of exogenous folates. By interfering with folate salvage, pyrimethamine would increase the parasite's reliance on de novo synthesis and thus its susceptibility to sulphadoxine. In agreement with this idea, folate competed with pyrimethamine in the presence of concentrations of sulphadoxine just sufficient to block de novo synthesis . Furthermore, transgenic strains with slightly (PKDSII) and drastically (PKDSM2) reduced DHPS activity relative to the parent strain FCB exhibited reduced (PKDSII) and abolished (PKDSM2) synergism between sulphadoxine and pyrimethamine . Questions remain as to the molecular basis for the pyrimethamine effect on folate utilisation and the applicability of this model to other antifolate combinations.
5.3Interactions among endoperoxides (artemisinins and related agents)
Artemisinin is believed to react via its endoperoxide group with haem in P. falciparum digestive vacuoles, resulting in the formation of free radicals that alkylate parasite proteins . The artemisinin series of compounds are all expected to interact with the parasite in much the same way so as expected no synergism or antagonism between them has been reported as far as the author is aware. However, the same does not necessarily apply to other endoperoxides: the trioxane Fenozan B07 was found to be synergistic with artesunate against a mouse infection with P. berghei and the tetraoxane WR 148999 was synergistic with artemisinin in cultured P. falciparum. The mechanistic basis of these observations is unclear and there has been little discussion of combining an artemisinin with another endoperoxide in the clinic.
5.4Interactions between quinoline and related agents and endoperoxides
There are several reports of synergism between artemisinins and quinine or mefloquine (Tables 1 and 2) and this could be a result of the different but related actions of the two groups on haem in the digestive vacuole. Mefloquine is being increasingly combined with artemisinins in the clinic, especially in south-east Asia where drug resistance is particularly prevalent . By contrast, there are reports of strain-specific antagonism between artemisinins and chloroquine (Table 1).
5.5Interactions between quinoline and related agents and other compounds
Khan et al.  showed clear antagonism between chloroquine and erythromycin for six out of eight cultured P. falciparum strains from the Kenyan coast region, in contrast to the synergism previously reported (see ). The contrast with previous results was put down to the longer test times (up to 7 days) used by Khan et al. This has clinical relevance as the two agents have been combined. Similar results were obtained with amodiaquine + erythromycin . A combination of quinine with rifampicin in a clinical trial of uncomplicated malaria was also unfavourable. The study showed shorter parasite clearance times for the combination than for quinine alone, but recrudescence rates were approximately five times higher for the former . This was attributed to the increased conversion of quinine to its less potent 3-hydroxy derivative in the patients receiving rifampicin.
The antagonism between chloroquine (or mefloquine) and the protease inhibitor Ro 40-4388 was explained by Sullivan et al.  as the result of the Ro 40-4388's inhibitory action on globin breakdown leading to reduced release of haem in the digestive vacuole and thus reduced quinoline–haem binding. This would make sense in the context of a toxic quinoline–haem complex and/or haem being a driver of chloroquine uptake.
The proton pump inhibitor omeprazole was slightly but significantly synergistic with quinine and antagonistic to chloroquine . It was suggested that the latter observation could be the result of reduced acidification of the digestive vacuole and consequently reduced accumulation of chloroquine, a weak base, in that compartment, its probable site of action. However, this suggestion could conflict with the more recent findings of Roepe and co-worker  that certain chloroquine-resistant strains actually had more acidic vacuoles.
5.6Interactions between antifolates and other compounds
The hydroxynaphthoquinone atovaquone (566C80) is the only completely novel antimalarial agent to reach the market in several decades . Early treatment failures with atovaquone monotherapy led Wellcome scientists  to test for interactions with a number of agents (see also below). Proguanil showed strong synergism with atovaquone, subsequently confirmed by others in almost all strains tested (Table 1), and a fixed-dose atovaquone + proguanil combination is now marketed. However, this author could find no published evidence that this combination is synergistic in vivo. Interestingly, although proguanil is extensively metabolised to cycloguanil in vivo and the latter is much more potent than proguanil, there is no consistent pattern of synergism between atovaquone and cycloguanil (Table 1). Moreover, patients carrying cycloguanil-resistant strains with mutations in the dhfr gene can often still be cured with atovaquone + proguanil . This suggests that the synergism with atovaquone is particular to unmetabolised proguanil. Srivastava and Vaidya  have proposed a mechanism of atovaquone–proguanil synergism in which proguanil enhances the collapse of the electrical component of the potential across the parasite's inner mitochondrial membrane (Δψm) caused by atovaquone. However, this effect is only seen at concentrations of proguanil close to those that on their own cause collapse of Δψm and it is not clear whether the effect is synergistic or additive.
5.7Interactions between endoperoxides and other compounds
Benoit-Vical et al. [34,35] combined artemisinins with metalloporphyrins with the intention that the latter, like endogenous haem, might activate the peroxide function of artemisinin in the parasite's digestive vacuole. The results were somewhat variable but synergism between artemisinin or artemether and one of the metalloporphyrins, meso-tetrakis(4-sulphonatophenylporphyrin) [TPPS] was demonstrated in cultured parasites [34,35]. However, the opposite effect was observed between artemether and TPPS in mouse malaria , which illustrates that results obtained with cultured parasites are not always replicated in vivo. The combinations of artesunate with the dicatecholate iron chelator FR160  or with any of the three hydroxypyridinone iron chelators deferiprone (CP20), CP38 or CP110  were antagonistic, presumably as a consequence of the role of iron in the action of artesunate. The same result was found for artemisinin plus desferrioxamine by Eckstein-Ludwig et al. . They proposed a specific target for Fe2+-activated artemisinin, namely PfATP6, a sarcoplasmic/endoplasmic reticulum Ca2+-ATPase (SERCA). The Ca2+-ATPase activity of recombinant PfATP6 in vitro was inhibited by artemisinin and this inhibition was antagonised by desferrioxamine, mirroring the result obtained using intact cells. The effect of artemisinin on cultured parasites was also antagonised by the SERCA inhibitor thapsigargin, but as discussed above, this does not necessarily provide evidence for a shared target.
5.8Interactions among protease inhibitors
Proteases of malarial parasites are important in haemoglobin degradation, invasion of erythrocytes and merozoite release, and protease inhibitors have received much attention recently as potential antimalarial drugs . Strong synergism between cysteine and aspartic endoprotease inhibitors (e.g., E-64 and pepstatin A, respectively) in culture was first demonstrated by Bailly et al.  and later extended to mouse malaria by Semenov et al. . The synergism was said to be the result of the apparent cooperative roles of the falcipain cysteine proteases and plasmepsin aspartic proteases in globin digestion in the digestive vacuole, but why the effect should not be greater than additive is not clear. Gavigan et al.  showed that the aminopeptidase inhibitor bestatin was also synergistic with both cysteine and aspartic protease inhibitors, albeit not so strongly. Aminopeptidase is believed to be responsible for the terminal stages of globin degradation to amino acids in the parasite cytosol .
One possible example of formation of a complex of enhanced toxicity as a basis for synergism is the interaction between Cu2+ and the Cu2+-chelator diethylthiocarbamate . Parasite membranes are suggested to be the target of this complex but this idea remains to be proven.
Plasmodium is particularly susceptible to pro-oxidant agents. Winter et al.  unexpectedly observed very strong synergism between the related pro-oxidants rufigallol and exifone. It was hypothesised that interaction between rufigallol and iron in the parasite's digestive vacuole led to the formation of hydrogen peroxide and subsequently hydroxyl radicals, which transformed the relatively inactive exifone into a more-potent xanthone. In support of this idea, the xanthone could be formed in vitro under mildly acidic conditions in the presence of iron, and this compound in an acetylated form was a highly potent antimalarial agent .
Salicylhydroxamate (SHAM) inhibits oxygen consumption via an alternative oxidase pathway and was therefore expected by Murphy and Lang-Unnasch  to complement the activity of atovaquone on the cytochrome chain. Both SHAM and propyl gallate, another alternative oxidase inhibitor, were indeed synergistic with atovaquone on one P. falciparum strain. The use of this alternative pathway has been suggested to provide sufficient bypass in atovaquone-treated parasites to allow some growth and emergence of atovaquone-resistance by virtue of mutations in the cytochrome b gene. Proguanil (see above) did not have the same effect on oxygen consumption as SHAM so this is not proposed to be the basis of the former's synergism with atovaquone .
A clinical antimalarial candidate, fosmidomycin, is the first to have arisen out of genomic studies of P. falciparum. Data mining of the (at that time incomplete) P. falciparum 3D7 genome sequence revealed the presence of genes encoding enzymes of the non-mevalonate pathway of isoprenoid biosynthesis . Fosmidomycin, an inhibitor of an enzyme deoxyxylulose-5-phosphate reductoisomerase, had potent antimalarial activity in culture and in vivo and was selective because mammals use a different (mevalonate) pathway. In view of the high level of recrudescences in clinical trials with fosmidomycin alone, Wiesner et al.  looked for a partner drug much as Canfield et al.  had for atovaquone. Clindamycin was shown to be somewhat synergistic with fosmidomycin against at least one strain in culture , but it would be desirable to see this observation extended to other strains and to assessment of interaction in vivo. Nonetheless, the combination was reported to be entering phase-II clinical trials.
The bis-quaternary ammonium compound T16 was designed as an inhibitor of phosphatidylcholine metabolism but also has an interaction with haem that is suggested to be relevant to its antimalarial action . Binding to haem was proposed to be crucial to the (several 100-fold) accumulation of T16 inside parasitised erythrocytes. The protease inhibitor Ro 40-4388, which reduces haem release from haemoglobin, reduced T16 binding to parasitised erythrocytes. In support of this idea, the two agents were antagonistic in their action on cultured P. falciparum. Similar arguments were made to explain the antagonism of Ro 40-4388 to the other haem-binding agents chloroquine (see above) and pentamidine .
Synergism and antagonism between antimalarial agents have been reported frequently but shortcomings in the analysis of some of the data make it difficult at times to assess whether the alleged interaction is genuine or not. This task would be made easier if the raw data were available. In the era of internet web sites and ‘on-line supplementary material’ this should be possible in future and ought to be considered for this type of data and other types where more than one analysis is possible. Other difficulties include different interactions in different species and strains and at different concentrations/ratios of drugs, and the use of different read-outs for growth in culture and different infection and dosing regimens and efficacy criteria in animal models. If standard procedures could be adopted, it would be easier to compare one laboratory's findings with another's and resolve some of the apparent contradictions. This will probably not happen before standard procedures for testing of single agents are more widely adopted.
While it is clear that confirmed, consistent synergistic and antagonistic interactions have implications for the mechanisms of action of the agents concerned, there has been in the literature considerable flexibility in the interpretation of results. In general, it is probably better to start with a prediction of synergism or antagonism based on a pre-existing hypothesis and then to test whether this prediction is correct. If the opposite approach is adopted, i.e., the observation is made then attempts are made to fit it into the hypothesis, there is a tendency for synergism/antagonism data to be subordinate to the researchers' preconceptions and to contribute little or nothing to the shape of the mechanistic model proposed.
The importance of antimalarial drug interactions for clinical practice is more obvious and studies of interactions should be done rigorously in culture and in vivo before a new combination is employed in malaria patients. Given the tendency towards combination therapy, determination of interactions between novel and established agents or combinations of novel agents will be crucial in the research and development of new antimalarial drugs. This assessment should probably be undertaken at the ‘secondary evaluation’ stage, when the new compound has already shown promising activity and selectivity as a single agent in culture and in vivo. Standardisation of strains, methods and analysis, as discussed above, would help to identify the worthiest combinations from the world-wide antimalarial drug ‘pipeline’ and accelerate the deployment of the new drugs that are so badly needed.
The author's recent antimalarial research has been funded by the Health Research Board, the Higher Education Authority Programme for Research in Third-Level Institutes (HEA PRTLI)-funded Institute for Information Technology and Advanced Computing (IITAC) programme, Enterprise Ireland, and the British Society for Antimicrobial Chemotherapy. He thanks Dr. Bill Watkins for comments on the manuscript.
The term ‘drug’ here includes experimental antimalarial compounds but compounds much too toxic to be conceived of as human medicines (e.g., paraquat ) are not discussed.
Resistance-reversing agents (such as the chloroquine-resistance reversers verapamil and trifluoperazine) are a special case in that synergism occurs in principle only in resistant strains. The same rules for analysis of interaction data apply in these cases, too, but they are excluded here because of space limitations.