Discussion
Broad inclusion criteria allowed most studies that report the frequency of
dhfr and/or
dhps genotypes in Africa to be included in this meta-analysis. The studies were conducted as early as 1993 and as late as 2006; however, the distribution is skewed (mean year: ~2001 for studies reporting either multiple mutant genotype) towards recent years reflecting an increase in the number of studies reporting the frequency of the
dhfr triple mutant and/or
dhps double mutant genotypes. This observation is in line with the increase in the number of anti-malarial clinical efficacy trials reported in the literature in recent years [
77] and likely reflects the growing prioritization and feasibility of surveillance as a component of malaria control strategies. 161 studies assembled in this meta-analysis collectively examined blood isolates derived from over 14,000 individuals for the
dhfr triple mutant genotype and nearly 8,000 individuals for the
dhps double mutant genotype. The clear emphasis on reporting the
dhfr triple mutant genotype is likely due to its emergence some 10-20 years earlier than the
dhps double mutant genotype in Africa [
78] and its ubiquitous presence across the continent. The
dhps double mutant genotype (437G/540E), on the other hand, emerged much later in the early to mid-1990's [
79] and is only prevalent in East Africa with a small number of cases reported in West Africa. The 436A/437G double mutant appears to be emerging in West Africa with apparent absence of 437G/540E double mutant genotype in some countries. The prevalence of individual 436A and 437G alleles have been summarized (see Additional File
1). However, the 436A/437G double mutant genotype was not independently summarized due to limited reporting of data on this genotype.
The maps describing the prevalence of the
dhfr triple mutant genotype (Figure
1) and
dhps double mutant genotype (Figure
2) illustrate a high concentration of studies in certain parts of the continent. This observation clearly demands an assessment of potential sources of reporting bias. A wealth of data in particular areas is due in part to the availability of research capacity in these areas to facilitate such studies. However, it is also due to molecular surveillance studies conducted more recently in some countries that provide a vast amount of data compared to traditional molecular studies that complement clinical efficacy trials. In fact, a majority of the multiple mutant genotype data available from Mozambique, Uganda, and Cameroon comes from just a few surveillance studies [
80‐
83] that examined a large number of isolates from many sites within each country. Surveillance studies in Kenya [
84] and Tanzania [
63,
85] also contributed a significant amount of molecular data relative to traditional molecular studies. Potential reporting and publication bias must also be considered when examining the snapshots of resistance hotspots in Figures
1 and
2. Areas where studies are concentrated appear to coincide with areas where there has been a higher reported frequency of the multiple mutant genotype. This may be due in part to bias in which molecular studies that show a high prevalence of mutant genotypes are more likely to be published than those that show little or no prevalence. Another potential explanation is that traditional molecular studies often complement clinical efficacy trials and are more likely to be conducted in areas where therapeutic failure rates are high and there is greater interest in the role resistance mechanisms may be playing.
This meta-analysis shows that there are some important differences in the reported frequency of the
dhfr triple mutant genotype and
dhps double mutant genotype. Specifically, the
dhfr triple mutant genotype appears to be found in high prevalence in areas throughout the continent whereas the
dhps double mutant genotype only appears to be prevalent in parts of East Africa. This observation corroborates findings published by Pearce
et al, which also illustrated high prevalence of the
dhps mutant genotype consisting of A437G, K540E in East Africa and low prevalence in West Africa. A number of factors such as the early emergence of the
dhfr triple mutant genotype in Africa [
78], pervasive drug pressure from the distribution of pyrimethamine salts for malaria prophylaxis in various parts of the world [
86], and human migration patterns [
87], may explain the greater distribution of this genotype throughout the continent. Another important difference between the
dhfr triple mutant genotype and
dhps double mutant genotype is the impact of varying reporting conventions for mixed genotype data. This study establishes that reporting conventions for mixed genotypes has a significant impact on the reported frequency of the
dhfr triple mutant genotype. However, this was not observed for the reported frequency of the
dhps double mutant genotype. This may be explained by differences in diversity for the
dhfr and
dhps genotypes found in Africa. Parasite populations that have greater diversity for a particular gene (ie. more than one genotype in high prevalence circulating in the population) are more likely to yield mixed genotypes than parasite populations that have a low diversity for a particular gene. Thus, differences in the way mixed genotypes are reported have a more pronounced effect on the prevalence rates reported for a gene that has greater diversity in a given population than one that does not. It has been shown that the
dhps (A437G, K540E) mutant genotype is either ubiquitously present or absent in most of the populations where it has been examined [
79]; therefore, polyclonal infections containing the
dhps double mutant genotypes along with another
dhps mutant genotype are less likely to occur. However, the
dhfr triple mutant genotype has historically been found alongside a higher prevalence of sensitive, single, and double mutant
dhfr genotypes circulating in the population [
87,
88]; therefore, mixed genotypes are more likely to occur, particularly in earlier studies before the triple mutant
dhfr genotype became fixed in the population. This phenomenon may explain why reporting conventions for mixed genotype data was significantly associated with the reported frequency of the
dhfr triple mutant genotype but not the
dhps double mutant genotype.
The specific reasons responsible for the differential impact of reporting conventions for mixed genotype data on the reported frequency for either mutant genotype examined in this meta-analysis are difficult to identify and may also be an artifact of confounding variables that were not accounted for in the statistical regression used herein. Regardless, the fact remains that polyclonal infections commonly occur in many of the high transmission settings across Africa and the characterization of mixed genotypes is an issue that must be addressed. Novel methods for analyzing mixed infections [
89], multivariate
in vitro studies, and standardized interpretation of existing mixed genotype data [
32] have all been proposed as methods to better characterize mixed genotypes and present promising avenues for the future. Although there are a number of differences in the distribution and prevalence of the
dhfr triple mutant genotype and
dhps double mutant genotype, there are also some similarities - The prevalence of both alleles is increasing on average in Africa. On average, the age and clinical status of the sampled population are not significantly associated with the reported frequency of either mutant genotype. This is in contrast to smaller regional studies that have shown association between age, treatment failure rate, and the prevalence of genetic markers for resistance [
90,
91]. The observations from this meta-analysis that suggest there is not a significant association between these factors should not be seen as a contradiction to more localized studies; rather, they testify to limitations that are inherent in meta-analyses, particularly in identifying regional trends. These limitations must be specifically addressed in order to contextualize the observations made in this study and appropriately articulate the conclusions that can be drawn from them.
One of the major limitations in this meta-analysis was the small number of variables that could be assessed in the statistical models. There are a number of clinical and epidemiological variables such as transmission, regional drug administration policies, and target populations (ie. pregnant women, co-infections, etc.) that may better explain the reported frequency of the dhfr triple mutant and dhps double mutant genotypes but could not be examined in this analysis. This was due to lack of power from limited datasets and inability to link these data to reported frequency data where this information was not specifically reported.
Another limiting factor that must be considered when making statements regarding trends across the continent is the confounding effect of space and time. The studies reporting the frequency of the
dhfr and
dhps multiple mutant genotypes included in this analysis were conducted over a period of 13 years. In most cases, the data from a particular country only spans a fraction of this 13 year period; therefore, there is a regional bias where data from a particular year is really a reflection of certain regions, not the continent as a whole. The sample size of studies included in the analysis may have also contributed to regional and temporal bias since the regressions were weighted. The mean study size was 90 isolates for studies reporting the
dhfr triple mutant genotype and 85 isolates for studies reporting the
dhps double mutant genotype. However, studies examined as few as 3 isolates and as many as 453 isolates (approximately 5% of the total number of studies had sample sizes of fewer than 15 isolates). Since studies that have small sample sizes are more likely to be skewed by rare events [
92], and these smaller studies tend to occur in more remote areas, this means that remote areas may not be as accurately represented by reported frequency data. In light of these factors, it is more appropriate to consider the data conveyed by the maps and regressions as snapshots of cumulative mutant genotype data rather than a description of mutant genotype frequency at any specific point in space or time.
Finally, accessibility to studies and the interpretation of data from included studies that use different reporting conventions represented noteworthy sources of bias and error, respectively. The literature search in this study was limited to the Pubmed database and Cochrane Library. Molecular studies that were not available through these sources are not reported. Studies that were included still required a level of interpretation due to differences in methodological and reporting conventions; some of which were specifically examined in this meta-analysis. However, many of these aspects simply could not be addressed within the scope this study. Since there is no established format for the collection and dissemination of molecular data, there was difficulty in determining if molecular data for the same samples was published multiple times in different articles. Efforts were made to filter out articles that reported data from isolates that were already described in a different article; however, there was uncertainty for some articles and it is possible that molecular data for some isolates were reported twice and others were excluded unnecessarily.
Conclusion
This database, despite its limitations, provides a centralized source for information describing the frequency of mutant dhfr and dhps genotypes in Africa that are known to confer anti-folate resistance. These data collectively show that the dhfr triple mutant genotype consisting of (N51I, C59R, S108N) is fixed in several parts of Africa and has been increasing, on average, over the time period from 1993 to 2006. The dhps double mutant genotype consisting of (A437G and K540E) also increased in prevalence, on average, during the same time period. However, this genotype appears to only be fixed in parts of East Africa while its prevalence in West Africa remains low. Continued drug pressure with SP may contribute to further selection of quadruple dhfr mutant genotypes and triple dhps mutant genotypes in Africa; similar to what has been observed in Asia and South America. The increasing prevalence and fixation of these mutant genotypes has historically been associated with a loss of anti-folate drug efficacy and may pose a threat to settings in Africa where anti-folates are still used to treat clinical malaria.
SP is currently used in sub-Saharan Africa for IPTp. In this context, the drug is used to clear
P. falciparum infection in women with acquired immunity and usually low density asymptomatic parasitaemia. Thus, the treatment efficacy of SP for clinical malaria in young children cannot be directly extrapolated to its usefulness for IPTp [
93]. However, a recent study by Harrington et al. [
94] in Muheza, Tanzania shows that IPTp was associated with increased parasitemia and a higher degree of placental inflammation. The study also demonstrates selection of the resistance allele at
dhps codon 581 in women receiving IPTp. In a subsequent study involving young children from Tanzania, emergence of triple mutant
dhps allele with mutations in codons 437, 540 and 581 has been confirmed and this coincided with the high rate of therapeutic failure [
95]. In the same report, preliminary data reporting the spread of parasites with
dhps mutation at codon 581 to other parts of Africa has been presented. These findings suggest that the spread of the highly resistant triple mutant
dhps allele along with triple/quadruple mutant
dhfr alleles in different parts of Africa could further compromise the effectiveness of SP for the treatment of falciparum malaria and may have implications for IPTp. Therefore, systematic molecular surveillance studies coupled with assessment of drug effectiveness, including in pregnant women receiving IPTp, will be critical in the future to correlate molecular surveillance data with SP efficacy.
A continuation of efforts to conduct systematic molecular surveillance will further provide valuable information needed to construct future iterations of resistant genotype maps and databases; however, standardized methodology and reporting conventions are critical to improve the resolution and statistical power of centralized information systems. Challenges confronted during this meta-analysis suggest that patient specific data, standardized characterization of mixed genotypes, more specific date and locality descriptions, and better design and distribution of molecular studies are all critical objectives that must be pursued in order to fulfill the vision of the centralized resistance network proposed by WWARN.
Acknowledgements
The database assembled here was made possible through valuable contributions made by a large number of scientists and professionals in the malaria research community. The work of these individuals is highlighted in Additional File 3.
We specifically thank Jeff Henry and the Geospatial Research, Analysis, and Services Program (GRASP) team (Division of Health Studies, National Center for Environmental Health, CDC/ATSDR) and Allen Hightower (Division of Parasitic Diseases, Center for Global Health, CDC) for their critical input and support in the development of the resistant genotype frequency maps. We also thank Alex Rowe and Larry Slutsker (Division of Parasitic Diseases and Malaria, Center for Global Health, CDC) for comments and edits to drafts of this manuscript. SS and LMS were supported by the Association of Public Health Laboratories, Silver Spring, MD, USA through the Emerging Infectious Disease Fellowship program. SS and SKM received additional support through the Atlanta Research and Education Foundation, VAMC, Atlanta. Funding from the CDC Antimicrobial Resistance Working Group is also acknowledged.
The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the United States Centers for Disease Control & Prevention.
Competing interests
The authors declare that they have no competing interests
Authors' contributions
VU and JWB conceived the original format and objectives of the meta-analysis. KMH produced an initial version of the database and genotype frequency maps. LMS and SS developed search criteria, data extraction methods, and tabular design. SS extracted and compiled the frequency data examined in the meta-analysis. SKM performed all statistical analyses and worked with SS and VU to write the first draft of the manuscript. All authors have read and approved the final draft of the manuscript.