Background

The potential expansion of artemisinin-resistant Plasmodium falciparum, from Asia to Africa, has heightened interest in identifying factors affecting resistance allele dynamics, including the re-emergence of drug-sensitive malaria parasites. Chloroquine and sulfadoxine–pyrimethamine (SP) were both once used as safe and effective primary treatment for uncomplicated malaria. Successive waves of anti-malarial resistance from Southeast Asia prompted public health organizations to abandon chloroquine, and then SP, in favour of artemisinin-based combination therapy (ACT).

After the removal of chloroquine drug pressure in Malawi in 1993, chloroquine-sensitive parasites re-expanded in the population, eventually outcompeting the chloroquine-resistant strains [1]. The return of chloroquine sensitivity was shown, via the analysis of selective sweeps, to be the result of the re-emergence of genetically diverse sensitive parasites that had survived selective pressure [2]. This relatively rapid and nationwide re-emergence of genetically diverse drug-sensitive parasites in Malawi was unexpected, as drug resistance had persisted in South America and Southeast Asia many years after reduction in drug pressure [1, 3, 4].

In late 2007, Malawi replaced SP with an ACT in response to failing SP efficacy. The persistence of the highly resistant haplotypes of dihydrofolate reductase (DHFR) and dihydropteroate synthase (DHPS), DHFR 51I/59R/108N and DHPS 437G/540E, 5 years after the reduction of SP drug pressure, was recently demonstrated by the authors, with discussion of possible causes and implications of continued resistance [5]. Selective sweep analysis suggested little to no fitness cost of SP resistance in the urban setting of Ndirande where local transmission intensity is presumed relatively low.

The effect of reduction of SP use on SP-resistant parasites has been variable in other regions. While the persistence of SP-resistant parasites in Venezuela was shown years after the removal of drug pressure [6], a study of the frequency of Peruvian DHFR and DHPS resistance haplotypes showed a decline in DHFR 51I/108N/164L haplotypes within 5 years of SP removal as well as a decline in DHPS 437G/540E/581E [7]. Another recent study in Kenya observed differential change in genotype prevalence between rural-moderate and rural-high transmission settings 2 years after replacing SP with an ACT [8]. It is possible that these differences could be due to different background rates of malaria transmission.

Malaria parasites from areas of high transmission tend to be more diverse and undergo more recombination [6, 9, 10]. In the absence of selective pressure, high parasite diversity and recombination rates begin to increase the level of heterozygosity observed around resistance loci [6, 11]. Recombination between resistant and sensitive parasites from diverse lineages will begin to increase heterozygosity within the resistant parasite population, decreasing the width and depth of the selective sweep [6, 11]. Higher malaria transmission also results in a larger proportion of clinically immune hosts that can serve as a reservoir of drug-sensitive parasites that can re-expand after removal of drug pressure [12]. Areas with low malaria transmission tend to have lower levels of parasite diversity, leading to higher rates of inbreeding and more rapid rates of fixation of polymorphisms, possibly explaining why re-emergence of sensitive parasites was not seen in Venezuela. The rate of malaria transmission throughout Malawi is higher than that observed in South America and Southeast Asia, yet relatively higher and lower transmission settings can be found in rural and urban environments within Malawi. In this study the authors estimate the prevalence of DHFR and DHPS resistance haplotypes and the characteristics of the associated selective sweeps at three sites with low, moderate and high parasite prevalence within Malawi to explore whether local variation in transmission impacts the re-expansion of sensitive parasites.

Methods

Samples were collected as part of a malaria surveillance study conducted in health facilities in three districts in Malawi: a rural site near a river with high parasite prevalence (Chikwawa), a rural more arid site with moderate parasite prevalence (Thyolo) and an urban site with low parasite prevalence (Ndirande). Samples were collected from individuals presenting to the hospital with uncomplicated malaria. All samples were collected with informed consent according to an institutional review board approved protocol. Samples consisted of blood spots, collected on filter papers, representing a patient’s initial infection on the day they were admitted to the study, prior to treatment. Parasite prevalence from a community-based, cross-sectional survey was used as a surrogate measure of transmission intensity; 8.4, 14.3 and 29.6 % of individuals surveyed in the urban-low, rural-moderate and rural-high settings, respectively, were qPCR-positive for P. falciparum parasites. Recent entomological inoculation rates (EIR) were not available for these sites, but the use of parasite prevalence as a surrogate for EIR and transmission intensity is supported in previous studies [13].

DNA was extracted from filter paper blood cards using a Qiagen BioRobot (Qiagen, Valencia, CA, USA) following the Investigator Bloodcard Protocol. Parasite genotypes at polymorphic sites within pfdhfr-ts and pfdhps genes were determined via pyrosequencing. Single nucleotide polymorphisms (SNPs) within codons 51, 59 and 108, of pfdhfr-ts and codons 437, 540 and 581 of pfdhps were genotyped for all samples using primers and amplification methods adapted from Zhou et al. [14]. Pyrosequencing was performed on a PyroMark Q96 MD system (Biotage, Charlotte, NC, USA). Allele frequency was adjusted based on a standard curve [15]. An allele with a relative frequency of 80 % or greater within a given infection was designated as the predominant allele. Haplotypes were constructed using only the predominant allele. Samples without a predominant allele at two or more codons were labelled ‘mixed-genotype’. Samples that were mixed at a single codon were treated as containing both possible haplotypes.

To show that reduced heterozygosity around drug resistance genes was the result of selection rather than demographic processes, expected heterozygosity was measured in six unlinked neutral loci located throughout the P. falciparum genome [16]. These unlinked microsatellites were amplified using previously published primers and amplification conditions [16]. When multiple peaks were identified within the same sample, peaks that were less than one-third the height of the tallest peak were ignored and the tallest peak was designated as the predominant allele [16, 17]. Only samples with a predominant allele were included in estimates of expected heterozygosity.

Eight polymorphic microsatellites flanking pfdhfr-ts were genotyped; four downstream (+50, +20, +1.48, +0.2 kb), and four upstream (−0.3, −1.2, −10, −30 kb) using previously described primers and protocols [11, 18]. Eight polymorphic microsatellites flanking pfdhps were genotyped; four downstream (+9.008, +1.407, +0.505, +0.034 kb), and four upstream (−0.132, −2.849, −7.489, −11.069 kb) of the gene, using primers described by Vinayak et al. [19]. Fragment size was visualized using an Applied BioSystems 3730XL high-throughput 96-capillary DNA sequencer. Analysis of electropherograms was performed using Genemapper software (version 4.0; ABI). A Perl script was used to assign the raw electropherogram scores to an integer allele size based on the expected repeat length and variation seen in the positive controls.

The prevalence of each haplotype was estimated as the number of each haplotype observed among the successfully genotyped samples for each resistance gene, divided by the total number of genotyped samples. Haplotype frequency, the proportion of parasites with a given haplotype, could not be calculated as the number of parasite clones per infection was not estimated. A resistant haplotype was defined as containing any number or combination of resistance alleles at the genotyped codons within either of the resistance genes of interest. The sensitive haplotype was defined as parasites with sensitive alleles at all codons within both genes. Samples with mixed-genotype at two or more loci within the same gene were excluded because haplotype phase could not be determined. Chi squared tests with Yate’s correction were used, where appropriate.

Expected heterozygosity (He), a measure of genetic diversity at each microsatellite locus, was calculated using the standard equations for He and variance:

$$H_{e} = \left( {\frac{n}{n - 1}} \right)\left( {1 - \sum {p_{i}^{2} } } \right),\quad \frac{2(n - 1)}{{n^{3} }}\left\{ {2(n - 2)\left[ {\sum {p_{i}^{3} - \left( {\sum {p_{i}^{2} } } \right)} } \right]} \right\}$$

The analysis focused on sweep characteristics flanking DHFR 51I/59R/108N and DHPS 437G/540E due to limited prevalence of other haplotypes in all three settings. He (±1 standard deviation) was calculated for three groups: rural high, rural-moderate and urban-low sites, for both pfdhfr-ts and pfdhps genes. Samples without a predominant genotype or samples with missing data were excluded from expected heterozygosity calculations. Statistical significance was determined via permutation. Diversity ratios were calculated for Hurban-low/Hrural-high, Hurban-low/Hrural-moderate, and Hrural-moderate/Hrural-high. Calculations for He, standard deviation and permutations were conducted in R [20].

Results

Of the available samples, complete pfdhfr-ts haplotypes were assembled for 549 samples from the urban-low site, 726 samples from the rural-moderate site, and 660 samples from the rural-high site (Fig. 1a). Complete pfdhps haplotypes were assembled for 546 samples from the urban-low site, 733 samples from the rural-moderate site and 558 samples from the rural-high site (Fig. 1b). No difference in the prevalence of the highly resistant DHFR 51I/59R/108N or DHPS 437G/540E was found between the three study sites. Haplotype prevalence of the DHFR 51I/59R/108N haplotype was >95 % for all study sites. The prevalence of the highly resistant DHPS 437G/540E was also >95 % for all three sites. A greater prevalence of DHFR 59R/108N was found in the rural-moderate site, relative to the rural-high site (0 vs 2 %, Yate’s corrected p = 0.020), and was borderline different from the urban-low site (1 vs 2 %, p = 0.067). A single, sensitive parasite infection was identified in the rural-moderate site.

Fig. 1
figure 1

Sulfadoxine–pyrimethamine-resistant haplotype prevalence at three sites with different prevalence of malaria parasites. Haplotype prevalence at DHFR (a) and DHPS (b) at three sites within Malawi: a rural site with high parasite prevalence (Chikwawa), a rural site with moderate parasite prevalence (Thyolo), and an urban site with low parasite prevalence (Ndirande). Prevalence was calculated as the percentage of individuals with a given haplotype. Error bars represent 95 % confidence intervals

Full size image

A sub-set of samples was subjected to microsatellite analyses to examine selective sweep characteristics between study sites. Of the sub-set, complete microsatellite haplotypes were generated for the rural-moderate site (n = 15, pfdhfr-ts), (n = 21, pfdhps), rural-high site (n = 10, pfdhfr-ts), (n = 18, pfdhps), and urban-low site (n = 15, pfdhfr-ts) and (n = 20, pfdhps). Average He at unlinked microsatellites was 0.898. The proportion of samples scored as polyclonal based on unlinked microsatellites did not differ significantly between study sites (p = 0.611). None of the sites had He values that were consistently different from each other (Fig. 2). Significant changes in He, were seen between the rural-moderate site, and the rural-high site and urban-low site (p < 0.001) though no differences between the rural-high and urban-low sites were found. Analysis of flanking microsatellites indicated the presence of a core haplotype flanking pfdhfr-ts and pfdhps, likely of Southeast Asian origin, in all three sites (see Additional file 1). Lastly, microsatellite haplotypes flanking pfdhps found in the urban-low and rural-high sites were observed that were not present in the rural-moderate site, and microsatellite haplotypes found in the rural-moderate site not found in the other two locations (Fig. 3).

Fig. 2
figure 2

Sulfadoxine–pyrimethamine selective sweep characteristics at three sites with different prevalence of malaria parasites. Expected heterozygosity in microsatellite loci flanking DHFR 51I/59R/108N (a) and DHPS 437G/540E (b). Samples with missing data were excluded. Error bars represent ±1 standard deviation. Dashed lines represent genomic level, average He, based on unlinked loci. Alpha <0.05 asterisk indicates significant difference between two study sites based on permutation

Full size image
Fig. 3
figure 3

Proportion of unique versus common microsatellite haplotypes between study sites. Proportion of microsatellite haplotypes found between study sites

Full size image

Discussion

This study estimates the effect of local differences in malaria transmission, as inferred from parasite prevalence, on the prevalence of SP-resistant haplotypes and the characteristics of associated selective sweeps within Malawi. Despite marked differences in the parasite prevalence between the rural-high and urban low sites, no difference in haplotype prevalence or sweep characteristics was observed. Statistically significant differences between the prevalence of DHFR double mutants in the rural-moderate and rural-high sites are possibly a statistical artifact due to large sample size and are not considered clinically relevant.

This study also indicated similarity in sweep characteristics and some shared microsatellite haplotypes between the rural-high and urban-low sites, counter to what would be expected under the assumption that higher transmission would result in greater genetic diversity. These findings may indicate a shared parasite population between these locations. The divergence of some sweep characteristics and microsatellite haplotypes in the rural-moderate site compared to the other two sites, suggests that regional demographic events may affect sweep characteristics, without a significant effect on haplotype prevalence. Such events could have been the migration of SP-resistant haplotypes of diverse genetic background to the rural-moderate site (perhaps from neighbouring Mozambique).

The similarity in the proportion of samples scored as polyclonal between study sites may indicate similarity in the level of malaria transmission, not reflected in the parasite prevalence observed at each site. Recent EIRs were unavailable for these study sites; therefore, parasite prevalence was used as a surrogate for transmission intensity. If parasite prevalence was an accurate estimation of transmission intensity in this study, a larger proportion of polyclonal samples in the higher transmission sites would have been expected. This discrepancy could indicate that parasite prevalence is not a good surrogate for transmission intensity in this study. The estimation of transmission intensity used here was based on average parasite prevalence from a community-based, district-wide, cross-sectional survey, while the data-set used in this study was based on a facility-based survey. The cross-sectional data provide a single estimate of parasite prevalence at a given point in time, and may not reflect historical differences in transmission intensity between the sites. It is possible that individuals reporting to the district health centres in this study may have been from areas with more similar malaria transmission than the district-wide average parasite prevalence suggests. It is also possible that the almost three-fold difference in parasite prevalence between these sites is not sufficient to produce a measurable difference in the proportion of polyclonal infections, haplotype prevalence, or sweep characteristics.

The data also indicate a Southeast Asian origin of the Malawian pfdhfr-ts haplotype based on similarity in microsatellite repeat length between Malawian parasites and the V1S control strain. While, the Malawian pfdhps haplotype in this study differed at several markers from the V1S strain, core markers (those in close proximity to pfdhps) were similar to those published by Mita et al., and Alam et al. showing similar core haplotypes found in Malawi, Bangladesh, Thailand and Cambodia [21, 22]. The lack of published microsatellite data for reference strains makes direct comparison between publications difficult, but these data suggest a Southeast Asian origin of the pfdhps resistance allele in Malawi.

This research presents evidence that after nearly two decades of high SP resistance in Malawi, differences in regional malaria transmission, as inferred from parasite prevalence, do not impact the prevalence of SP-resistant parasite haplotypes. The SP-resistant haplotypes, DHFR 51I/59R/108N and DHPS 437G/540E, have maintained their high prevalence long after the removal of SP as the first-line treatment in multiple settings in Malawi. Investigators working in epidemiologically similar settings can reasonably assume similarity in SP-resistant haplotype prevalence and sweep characteristics between local sites with different parasite prevalence.