The first-line anti-malarial treatment regimen for uncomplicated malaria in Kenya changed from CQ to SP in 1998 and from SP to AL in 2004. In both instances, the change was necessitated by widespread treatment failure. The prevalence of SP and chloroquine resistance-associated genotypes and in vitro resistance was compared between the period just after SP implementation (1999-2000) and a period five to seven years later (2003-2005) in patients with febrile illness presenting to outpatient clinics in Kisumu, Kenya. Malaria in Kisumu and in other areas surrounding Lake Victoria is endemic. For the 12 months ending in June 2004, the entomological inoculation rate in Kisumu district was 31.1 infective bites per person per year , and in the time period 2003-2006, the median P. falciparum parasite rate for children 2 to 10 years of age (PfPR2-10) was 71% in Kisumu . Despite changes in government anti-malarial policy, as late as 2010 SP was used to treat malaria in 37% of households surveyed in Kisumu, compared to 32% that used ACT .
The prevalences of pfdhfr codons N51I, C59R, and S108N were near saturation in the baseline study  and remained relatively stable between the baseline and follow-up studies. In contrast, the prevalence of mutations at all five pfdhps codons analysed increased dramatically over the same period. It thus appears that a significant proportion of isolates carried pfdhfr mutations prior to 1998, and use of SP as the first-line anti-malarial in Kenya resulted in the development of highly mutant pfdhps. Pre-existing mutations in pfdhfr may have derived from a combination of two sources: the use of SP as a second-line anti-malarial in Kenya prior to 1998 and the ongoing treatment of persons with HIV/AIDS with co-trimoxazole, a bacterial DHFR/DHPS inhibitor used to treat respiratory tract infections and prevent opportunistic infections. Co-trimoxazole shows cross-resistance with pyrimethamine and sulphadoxine in in vitro P. falciparum culture [49, 50] and may play a role in the development of mutations in pfdhfr and pfdhps.
One goal of this study was to determine whether the prevalence of the clinically relevant pfdhfr/pfdhps quintuple mutant, which is associated with SP treatment failure in Africa [13, 51], changed during the course of SP use as the first-line anti-malarial in Kenya. Between the baseline and follow-up studies, the prevalence of the quintuple mutant doubled to 53.5% and may be even higher. Due to absence of MOI data in these studies, pfdhfr/pfdhps haplotypes were deduced according to the method described by Kublin and colleagues . Because this method conservatively classifies isolates that are pure mutant at one pfdhfr allele and mixed at two other pfdhfr alleles as double mutant, the prevalence of the pfdhfr triple mutant may be systematically underestimated in populations where mixed infections are common. In the follow-up study, many more isolates had mixed pfdhfr genotypes than at baseline, and the prevalence of the pfdhfr triple mutant decreased between the baseline and follow-up studies. This observation contrasts with other findings showing that the population of SP-resistant parasites expanded over this period, and may be explained by more severe systematic underestimation in the follow-up study compared to the baseline. Thus, the prevalence of the pfdhfr triple mutant and the pfdhfr/pfdhps quintuple mutant in the follow-up study may actually be higher than estimated here.
Although the prevalence of the pfdhfr triple mutant appeared to decrease over time, increases in the prevalence of the pfdhps A437G/K540E double mutant led to a rise in the pfdhfr/pfdhps quintuple mutant prevalence. The very high prevalence of mutant codons 437 and 540 is consistent with other reports that these mutations are common in western Kenya [19, 52]. Notably, the data presented here show that SP use in Kenya was not just associated with the expansion of A437G and K540E, but also the progressive accumulation of mutations in pfdhps at codons 436, 581, and 613. This process appears to be rapid, as demonstrated by the emergence and spread of the A581G and A613S/T mutations in the short time between the baseline and follow-up studies.
Mutations at pfdhps codons 436, 581, and 613 are associated with increased levels of in vitro resistance when they occur with the A437G mutation [20, 43]. Among isolates in the follow-up study, over half had the mutations S436A/F, A581G, and A613S/T in addition to the A437G/K540E single or double mutant; A437G was present in either mixed or pure form in all single mutant isolates. Thus, at follow-up the majority of isolates had parasite genotypes associated with extremely high in vitro sulphadoxine resistance. Although the effect of these mutations in vivo is not well-studied, under SP selection, parasites containing the quintuple mutant and A581G have been found to have a selective advantage in vivo over quintuple mutant parasites . This selective advantage is reflected in the swift development and expansion of pfdhps mutations between the baseline and follow-up studies.
This is the first study to report the presence of the A581G mutation in Kenya. Although relatively rare in sub-Saharan Africa, this mutation has been documented in the neighboring countries of Uganda and Tanzania . In the background of mutant pfdhfr in Tanzania, a similar increase in the prevalence of A437G and K540E accompanied by the emergence and rapid spread of A581G was observed over the six year period following SP implementation . Notably, in this and other East African studies that reported the presence of A581G, all or most samples were wild-type at codon 613 [46, 53, 54]. This contrasts sharply with the findings reported here, where mutant codon 613 was present in over half of patient isolates in the follow-up study. It thus appears that the mutation A581G is becoming increasingly common in East Africa, while Kisumu may be a regional hotspot for the mutation A613S/T.
Although progressive mutations in pfdhps were observed, the pfdhfr mutation I164L, which is associated with high levels of chlorcycloguanil and pyrimethamine resistance in South America and South East Asia, was absent from isolates in both study periods. This mutation was detected in isolates from Kisumu in 2002 , prior to the start of the follow-up study, and in western Kenya in 2004 . The absence of the I164L mutation in isolates from the follow-up study, conducted from 2003-2005, indicates that parasites harboring this mutation in 2002 did not become widespread in Kisumu. Taken together, these findings are consistent with evidence that the I164L mutation is rare in sub-Saharan Africa, outside local hotspots in south-west Uganda and Rwanda [16, 19, 55].
The change from CQ to SP as the first-line anti-malarial in 1998 was expected to decrease CQ use in Kenya, possibly leading to restoration of CQ sensitive parasites. In Malawi, discontinuing the use of CQ led to the reversion of codon 76 to wild-type over a seven year period, and CQ sensitivity in vivo was restored after 12 years [32, 33]. Trends of increasing prevalence of wild-type codon 76 have recently been reported in other areas of East Africa. In Tanzania, the prevalence of wild-type pfcrt codon 76 increased over a six year period after anti-malarial policy replaced CQ with SP . Surveillance in coastal Kenya from 1993-2006 showed that the wild-type codon 76 prevalence also increased, although at a slower rate than in Malawi . Therefore, it was surprising to observe a statistically significant increase in the prevalence of isolates containing pfcrt K76T from 82% to 97%, as well as a significant increase in the prevalence of pfmdr1 N86Y, also associated with chloroquine resistance, from 44% to 76%. To analyse whether the apparent increase in the prevalence of mutant codon 76 arose from the higher prevalence of polyclonal infections observed in the follow-up study, the method of Hill and Babiker was used to estimate the frequency of the K76T allele. Even accounting for polyclonal infections, there was an apparent although not statistically significant increase in the frequency of K76T. Thus, the K76T mutation does not appear to be abating in Kisumu, as it is in coastal Kenya and in other areas in East Africa.
The persistence of the pfcrt K76T mutation in isolates from a high transmission setting such as Kisumu indicates that selection for the mutant codon is ongoing. The pfcrt K76T mutation is associated with amodiaquine as well as chloroquine treatment failure . Although AQ was officially the second-line therapy used in Kenya during the study period, it was available at 95% of drug retail outlets, whereas SP was available at only 29% of outlets . By comparison, CQ was stocked by 15% of drug retail outlets . It thus appears that widespread use of AQ in the private sector may be high enough to exert selective pressure on the parasite population.
Specific pfmdr1 genotypes have been linked to altered susceptibility to artemisinin, a component of the current first-line anti-malarial therapy AL. In vitro, the allele combination S1034C/N1042D/D1246Y is associated with increased artemisinin susceptibility , and in vivo, treatment with AL selects for the N86, Y184F, and D1246 alleles [29, 30]. In this study, the prevalence of alleles associated with artemisinin sensitivity (S1034C and D1246Y) decreased significantly; however, this decline is not likely due to selective pressure induced by AL as completion of the follow-up study predated widespread use of AL. Whether the prevalence of pfmdr1 mutations continues to change under increasing AL pressure will be of interest.