All participants provided written or oral informed consent. Ethical approval for project activities was provided by the review boards of the Malawi Health Sciences Research Committee, the University of Malawi College of Medicine Research Ethics Committee, the Liverpool School of Tropical Medicine, Macro International, the School of Public Health of the University of Kinshasa, the International Clinical Studies Review Committee of the National Institutes of Health, the Seattle Biomedical Research Institute, the Tanzanian National Institute for Medical Research, and the University of North Carolina.

Parasites from Malawi were obtained from peripheral blood of women who delivered children at Queen Elizabeth Central Hospital in Blantyre, Malawi, during 1997–2005 (18). In 2010, consecutive women who delivered children at study sites near Blantyre were offered enrollment into an observational study (F.O. ter Kuile et al., unpub. data). Dried blood spots were prepared from maternal peripheral and placental blood of enrollees.

Parasites from the DRC were obtained from adults in the 2007 Demographic and Health Survey (19). Parasites from Tanzania were obtained from placental blood of pregnant women delivering at Muheza Designated District Hospital during 2002–2005 (12).

For parasites from Malawi and DRC, genomic DNA was extracted from dried blood spots by using Chelex-100 or a PureLink 96 DNA Kit (Life Technologies, Grand Island, NY, USA), and P. falciparum was detected by using real-time PCR (19). These parasites were genotyped at dhps loci by using amplification and Sanger sequencing (18,20), and only those with pure A581G genotypes were genotyped at microsatellites. For parasites from Tanzania, the mutant alleles A437G and K540E are nearly fixed; A581G was identified by pyrosequencing (12). We classified parasites as having A581G if the mutant allele frequency was ≥90% within the parasitemia level of the person.

Five microsatellite loci flanking the dhps gene were genotyped in all isolates: −2.9 and −0.13 kB upstream, and 0.03, 0.5, and 9 kB downstream of dhps (20). PCR products of amplifications of individual loci were sized on a 310 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA), and allele lengths were scored by using GeneMapper v4.1 (Applied Biosystems). In specimens with multiple peaks, the major peak was analyzed. All specimens were amplified and sized in parallel with genomic DNA from P. falciparum isolate 3D7 (American Type Culture Collection, Manassas, VA, USA). These controls were used to correct allele lengths to account for batch variability in fragment sizing.

We computed heterozygosity (H_e) of microsatellite loci by using GenAlEx v6.5 (21) to quantify the degree of selection on mutant haplotypes. To assess relatedness among dhps haplotypes in Malawi during 1997–2010, we used GenAlEx to compute Φ_PT by analysis of molecular variance (AMOVA) with 999 permutations over the whole population (22) and the Nei genetic distance (23) among dhps haplotypes and years based on microsatellite profiles. We inputted Φ_PT values computed by AMOVA into a principal coordinates analysis (PCoA) in GenAlEx.

We further characterized these relationships with a network analysis. To characterize these relationships, we assigned unique haplotypes based on microsatellite profiles for the 91 isolates for which we had successfully genotyped all microsatellite loci. These unique haplotypes were inputted into NETWORK v4.6.1.1 (24,25), and weights were assigned to each locus in inverse proportion to the H_e, of the locus, as calculated above.

In cross-sectional analysis of parasite populations from eastern Africa defined by location and dhps haplotype, we first used GenAlEx to compute pairwise linear genetic distances and Φ_PT (by using AMOVA with 999 permutations over the full population) and then used SPAGeDi v1.4 (26) to compute pairwise R_ST (by using jackknifing with 1,000 permutations). We inputted pairwise tri-distance matrices of linear genetic distance, Φ_PT, and R_ST into separate PCoAs in GenAlEx. For testing of statistical significance, we considered a p value of 0.05 as sufficient to reject the null hypothesis and used the Bonferroni correction when computing multiple comparisons.

We constructed a neighbor-joining (NJ) network to estimate a phylogeny of dhps haplotypes circulating in eastern Africa. To construct this network, we first computed pairwise linear genetic distances among all 193 isolates in GenAlEx; this distance matrix was used to compute an unrooted NJ tree in PHYLIP v3.67 (27,28), which was computed agnostic to dhps haplotype and geographic location and rendered in R v3.0.1. Missing alleles precluded computation of a median-joining network with NETWORK (Technical Appendix Table 1).

We investigated population structure of the dhps haplotypes from these 193 isolates by using STRUCTURE v2.3.4, a clustering algorithm designed to infer and assign individuals to subpopulations (29). Although it was not specifically designed to identify population structure based on linked loci, we used STRUCTURE to test our a priori hypothesis of distinct subpopulations based on dhps haplotype and location (30). We performed 3 analyses: first, of all 193 parasites of all dhps haplotypes; second, of 116 parasites with any mutant dhps haplotype; and third, of 32 parasites with only the triple-mutant SGEG dhps haplotype. We performed 5 simulations each at values of K-estimated populations from 1 to 20, and estimated the true K a posteriori by using estimations in STRUCTURE, as well as using the method of Evanno et al. (31)

We first tested 114 P. falciparum isolates from Malawi collected during 1997–2010. We identified 25 wild-type SAKA parasites, 1 single-mutant SGKA, 68 double-mutant SGEA, 10 triple-mutant SGEG, and 10 with other dhps haplotypes, including AAKA, AGEA, AGKA, and SAEA. Among major haplotypes, we observed reductions in microsatellite allele mean heterozygosity (H_e) in parasites having double-mutant SGEA (H_e 0.454, SE 0.076) and triple-mutant SGEG dhps haplotypes (H_e 0.485, SE 0.134) compared with those from wild-type parasites (H_e 0.798, SE 0.064). These findings are consistent with positive selection on mutant haplotypes, presumably caused by sulfadoxine pressure.

Given the recent emergence of triple-mutant SGEG haplotype in 2010, we investigated its relationship with the double-mutant SGEA haplotype that had become fixed in this population by 2005 (32). To quantify genetic relatedness among years and major dhps haplotypes, we first computed pairwise Φ_PT values and Nei genetic distance among major dhps haplotypes binned by year. In these analyses, SGEA haplotypes were closely related to each other during 1997–2005 (Φ_PT values 0.008–0.065, Nei value 0.016–0.073) and closely related to the SGEG haplotype that emerged by 2010 (Φ_PT 0.082, Nei value 0.049) (Table). We inputted Φ_PT estimates into a PCoA to better visualize divergence among haplotypes by year. In this analysis, coordinates 1 and 2 explained 96.3% of the variance; SGEA haplotypes from all years clustered with SGEG haplotypes from 2010, which suggested a shared lineage in Malawi of mutant dhps haplotypes during 1997–2010 (Technical Appendix Figure 1).

We further investigated this finding by using network analysis. To perform this analysis, we constructed a median-joining network of wild-type and mutant haplotypes by year based on microsatellite profiles (Figure 1). In this analysis, we observed clustering of triple-mutant SGEG haplotypes from 2010 in a network of double-mutant SGEA haplotypes, as well as substantial sharing of microsatellite profiles among SGEA parasites from different years and with SGEG parasites. These 2 observations suggest a shared lineage of evolved mutant dhps haplotypes in Malawi.

Clinical consequences of infections with parasites bearing the dhps A581G mutation appear to vary among study sites in Africa. In Tanzania, these parasites have been associated with exacerbation of placental inflammation in women who received IPTp (12) and failure of IPTp-SP to prevent low birthweight of infants (33). In Malawi, these phenomena have not yet been observed. Because of these differing effects, we speculated that haplotypes bearing the A581G mutation may also differ among sites.

We conducted a cross-sectional analysis of parasites from 2 additional cohorts: 1) adults sampled in 2007 from the eastern DRC (19), and 2) pregnant women who gave birth and were enrolled during 2002–2005 in Muheza, Tanzania (12). In total, we compared the genetic relationships among 193 parasites grouped into 7 parasite populations: wild-type (SAKA) isolates from the DRC (n = 53) and Malawi (n = 24), those bearing double-mutant (SGEA) haplotypes from the DRC (n = 17) and Malawi (n = 67), and those bearing triple-mutant (SGEG) haplotypes from the DRC (n = 5), Malawi (n = 10), and Tanzania (n = 17). Fragment lengths are shown in Technical Appendix Table 1.

We quantified divergence of these 7 populations based on microsatellite allele lengths by using 3 population genetic metrics: linear genetic distance, Φ_PT, and R_ST. Linear genetic distance is a simple Euclidean genetic distance metric, and Φ_PT, and R_ST are variations of Wright F-statistics that quantify divergence and estimate its variance (22,34). Among SGEG parasites from Malawi and Tanzania, Φ_PT (0.420, p = 0.001) and R_ST (0.436, p<0.0001) indicated significant divergence after Bonferroni correction for multiple comparisons (Technical Appendix Table 2); Φ_PT and R_ST for other pairwise comparisons among SGEG parasites from Malawi, the DRC, and Tanzania were not significant but suggested similar divergence (values >0.420).

We visualized the output of each of these metrics with separate PCoAs (Figure 2; Technical Appendix Figure 2). In the PCoAs of genetic distance, Φ_PT, and R_ST, the first 2 coordinates accounted for 89%, 94.3%, and 96% of variance in values, respectively. In each PCoA plot, SGEG parasites from Malawi, the DRC, and Tanzania were consistently distant from each other in the 2 plotted dimensions, and other relationships among populations were variable. These analyses suggested divergence of SGEG haplotypes in Malawi, Tanzania, and the DRC.

To further investigate this apparent divergence of SGEG haplotypes, we computed an NJ network based on pairwise linear genetic distances among all 193 isolates. This phylogenetic analysis, which was computed agnostic to country and dhps haplotype, clustered all 17 SGEG parasites from Tanzania distinctly from 8 of the 10 SGEG parasites from Malawi; haplotypes in these clusters were also distinct from 3 of the 5 SGEG parasites from the DRC (Figure 3). A large number of SGEA parasites from Malawi also clustered closely with SGEG parasites from Tanzania, which suggested some shared lineage. This inferred phylogeny further suggests that triple-mutant dhps haplotypes from Tanzania and Malawi bearing the A581G substitution have arisen independently.

To further investigate this partition of dhps lineages, we performed an analysis of population structure. The first analysis of all 193 wild-type and mutant parasites partitioned only dhps haplotypes into mutant and wild-type, irrespective of geographic location. When restricted to the 116 parasites with mutant dhps haplotypes, the program partitioned the dataset into 3 clusters with 99.9% posterior probability; these 3 clusters directly corresponded to geographic location but were irrespective of dhps haplotype, which suggested separate lineages of mutant dhps haplotypes in Malawi, the DRC, and Tanzania (Figure 4, panel A). In an analysis restricted to the 32 parasites bearing the SGEG dhps haplotype, 2 population clusters were identified with the highest posterior probability. All SGEG parasites from Tanzania were assigned to 1 cluster, and SGEG parasites from Malawi and the DRC were assigned to a second cluster (Figure 4, panel B). These analyses further suggest that lineages of triple-mutant dhps haplotypes bearing the A581G mutation from Malawi and Tanzania are divergent.