The primary objective of this study was to describe the prevalence of G6PD genotypes in African malaria patients who participated in two CDA Phase III clinical trials. In this study, G6PD*A- prevalence varied between trial centres and between countries (Table 1). Where comparable data exists for each country, the prevalence of G6PD deficiency in the current study was lower compared with previous reports from Burkina Faso (31.0%) , and Nigeria (21.6%, 23.9% and 24.2%) [22–24], but higher than reported for Ghana (8.5%) . A study in Mali of children with uncomplicated or complicated malaria, reported a G6PD*A- prevalence of 12.9% ; about the same as reported here (10.5%). There are no previous reports for G6PD genotype prevalence using DNA analysis for Kenya or Tanzania for comparison.
The current study selected for particular malaria patients, so it is not surprising that G6PD gene frequencies are divergent from those reported for randomly selected healthy individuals in the general population. Notably, patients with known G6PD deficiency or neonatal hyperbilirubinaemia were excluded. Unfortunately, data are not available on how many patients were excluded from the clinical trials because of known G6PD deficiency. However, anecdotally, there were probably very few as the regions studied were naïve to G6PD testing. Conversely, G6PD prevalence data from a general population cannot necessarily be used to estimate the proportion of G6PD-deficient individuals that will be enrolled into an anti-malarial clinical trial. From the postulated evolution of G6PD deficiency, it is possible that the frequency of malaria is lower in patients with G6PD deficiency, though a protective effect against uncomplicated malaria has not been conclusively demonstrated.
In this study, all the G6PD*A- mutations involved A376G/G202A, confirming this variant as the most common in Africa. The A542T, G680T or T968C G6PD*A- mutations were not detected in the malaria patients included in this study. Data on the frequency of these mutations are sparse in Africa. The 968C mutation has been reported as the most common G6PD*A- allele in The Gambia  and in the Sereer ethnic group from Senegal . Given the selected population in this study it cannot be confirmed whether any of these mutations are not present in the countries studied. The limitations of the assay regarding other possible G6PD-deficient mutations that were not tested for in this study should also be considered. Further genotyping studies are required to determine the frequency of these mutations in the general population and specific ethnic groups.
The Hardy-Weinberg equation describes a population in which both allele and genotype frequencies do not change, meaning they are in equilibrium. Using this equation, ten centres had female gene frequencies significantly different to those predicted from male gene frequencies (Table 1). The sampling methods in the studied population (as determined by the study inclusion/exclusion criteria) may not have enrolled the different genotypes at the same frequencies that they occur in the general population. Also, the sample size was not large enough in some centres to confirm or reject the hypothesis of genetic equilibrium. Other explanations include changes in selection pressure, such as from malaria control or anti-malarial treatment, or population mixing. Evidence from Taiwan indicates that recent immigration can cause changes in relative G6PD gene frequencies between males and females independent of the overall incidence of G6PD deficiency . This may be relevant for some of the centres in this study; for example, Eldoret is one of the fastest growing towns in Kenya, and with 250 different ethnic groups in the country, it would be unlikely if genetic equilibrium for G6PD genotypes was maintained. However, it is difficult to separate the complex reasons for the divergence from Hardy-Weinberg equilibrium and further specific investigations would be required. There is also a methodological explanation. In many studies on G6PD-deficiency variant prevalence, only the frequency of the A- allele is known and this alone is used to derive predicted female genotype frequencies for G6PD deficient, heterozygous and normal; thus, only the A- allele is tested for equilibrium. In this study, the frequencies of the A-, A and B alleles in males were used to calculate predicted frequencies of the female genotypes, A-A-, AA-, BA-, AA, BB and BA, which is a strict application of the Hardy-Weinberg equation and so less likely to show conformance unless all G6PD alleles and genotypes are in equilibrium.
G6PD genotyping has limitations, for instance, it is not practical as a routine test in a clinical setting. Also, because the gene is highly polymorphic (more than 400 reported variants), unusual but clinically important variants can be missed . Phenotyping using the NADPH fluorescence spot test is an alternative test as it detects deficiency irrespective of underlying gene mutations (Table 2). The biggest limitation of G6PD phenotyping is its inability to conclusively identify heterozygous females because their G6PD levels can range from near normal to deficient. As shown in Table 3, 73.6% (128/174) of female heterozygous patients were classified as 'normal' by G6PD phenotyping. In the CDA trials, heterozygous females appeared to have no greater risk for clinically significant haemolysis than normal females (haemoglobin decrease of ≥ 40 g/L or ≥ 40% versus baseline or haemoglobin < 50 g/L or blood transfusion) [10, 11]. However, this cannot be assumed for all drugs.
As shown in Table 3, the specificity of phenotype testing for identifying G6PD*A- was 66.1% (82/124), i.e. 33.9% (42/124) of G6PD*A- patients were not identified by phenotype. This misclassification is probably attributable to reticulocyte rebound subsequent to haemolytic crisis, or recovery from malaria . In patients with malaria, higher than expected G6PD enzyme levels can occur in individuals with G6PD-deficient genotypes because the increased erythrocyte replacement rate results in a younger erythrocyte population and newly formed erythrocytes have a greater capacity for G6PD production . However, for unknown reasons, the correlation between deficient phenotype and G6PD*A- genotype was better for hemizygous males, specificity 70.7% (70/99), than for homozygous females, specificity 48.0% (12/25). This has also been seen in a recent paper where 83.3% (30/36) of hemizygous males were phenotypically deficient whereas specificity was only 60.0% (3/5) in homozygous females, though numbers in the study were very low compared with the current dataset . In males and females with normal G6PD genotype (A, B, AA or AB), 4.3% (25/575) and 3.1% (14/446), respectively, were found to be phenotypically deficient. It is possible that these individuals might have had G6PD mutations at loci other than those examined.
In the context of this trial, using only phenotype data would have significantly undermined interpretation of the drug-induced G6PD-related haemolytic effect of CDA as about a third of G6PD*A- patients would be expected to have had a clinically important haemolysis [10, 11]. Because of this, and the limitation in identifying heterozygous females, G6PD genotyping will continue to be important, particularly in a research situation. In the wider context, it has been suggested that phenotype could be used to exclude G6PD*A- patients from receiving potentially harmful treatments. It is possible to calculate the number of G6PD*A- patients that would have been inadvertently treated after phenotype screening (Table 3). Excluding the 134 patients who had a deficient phenotype would leave a total population of 1185 exposed to treatment, 42 (3.5%) of whom were of G6PD*A- genotype. Excluding patients with a deficient or intermediate phenotype (n = 261), leaves a total population of 1058, of whom 16 (1.5%) had a G6PD*A- genotype. The number of G6PD*A- patients misclassified by phenotype appears small (1.5%), i.e. about five patients would have had significant hemolysis that would have been avoidable with genotyping. However, where alternative therapies exist, in the clinical setting it would be unethical to expose even a small proportion of patients with a G6PD-deficient genotype to potentially life-threatening haemolysis, particularly where patient follow up is limited.
Across all patients, phenotype correlated reasonably well with the occurrence of significant hemolysis. Of the 878 patients with phenotype data, significant hemolysis occurred in 21.1% (19/90) of those patients who were phenotypically deficient, in 5.6% (5/90) of those who were intermediate and in 1.1% (8/698) of those with normal phenotype. Of the 41/124 (33.0%) G6PD*A- patients treated with CDA who had clinically important haemolysis, 23 had phenotype data available. Of these, 17/23 (73.9%) were phenotypically deficient, 4/23 (17.4%) had intermediate phenotype and 2/23 (8.7%) had normal phenotype.
The study design excluded patients with haemoglobin concentrations ≤ 70 g/L and this consideration may have affected the study results. There were no independent effects of G6PD genotype on baseline haemoglobin, temperature, or malaria parasitaemia in the study population (Table 4, Figures 2, 3 and 4). Thus, judged by these parameters, G6PD-deficient malaria patients appeared to be no different from other malaria patients at presentation. These results are consistent with a study in Tanzania in malaria patients that showed no effect of G6PD genotype on haemoglobin concentration or parasite density . Also, in Nigerian subjects with no or asymptomatic malaria, G6PD deficiency had no significant effect on haemoglobin levels, though red blood cell concentrations were higher in G6PD-normal subjects versus those who were G6PD deficient .
As shown in Table 5, there was a trend for patients with G6PD heterozygous or deficient genotypes or G6PD deficient phenotype to have lower unadjusted reinfection rates. However, after adjustment for baseline factors, including drug treatment, no effect of G6PD genotype or phenotype on recrudescence or reinfection rates was found. A recent study found no effect of G6PD deficiency on P. falciparum clearance after treatment with artemisinin-based combination therapy . However, there is some evidence that G6PD deficiency may reduce the number of parasite strains carried by individuals in transmission zones/seasons .
There are limitations in performing post-hoc statistical analysis on the effect of G6PD status on baseline variables and efficacy as was done in this study. Logistic modelling also has limitations; inclusion of the variables for adjustment is somewhat subjective and can never be comprehensive. Therefore, the conclusions that can be drawn from this study should be viewed as exploratory and require verification in prospective studies.