Background
During the 1970s and 1980s, malaria incidence increased in South America due to the disorganized development of cities and the emergence of drug resistance in
Plasmodium falciparum [
1,
2]. However, during the last 10 years, transmission has decreased and many areas exhibit limited spatial connectivity between parasite populations [
3]. As a result of this tangible reduction in transmission, malaria control programmes in these areas are expected to change from the control to the elimination phase.
Previous studies of
P. falciparum in South America identified strong genetic structure and epidemic expansions that were the result of inbreeding and bottlenecks due to high drug pressure [
4‐
7]. As a result, parasite populations within this region exhibit high genetic differentiation. Regardless of these discoveries, there have been few attempts to incorporate information on parasite genetic diversity into control programmes [
8‐
12]. A recent example is the study by Daniels et al. in Senegal [
11]. This study illustrated how genetic information obtained by using SNPs on
P. falciparum samples collected over an 8 years period enriched epidemiological information [
11,
12]. In particular, the authors could detect genetic changes in the parasite population that correlated with field data by using epidemiological models [
11].
In this paper, the relationship between parasite genetic variation and malaria epidemiology is explored in a locality that has seen substantial fluctuations in malaria prevalence. This study was conducted on samples collected over 8 years in the northwest region of Colombia, one of the most malaria-endemic countries in South America [
13]. Due to economic activities, such as mining and agriculture, there is substantial migration of workers with very high exposure to malaria (in close proximity to
Anopheles sp. breeding sites) into different towns. These local movements are considered important in the maintenance of malaria transmission in Colombia and other similar settings in the Americas [
1,
14]. Samples were genotyped at several microsatellite loci and analysed using Bayesian coalescent methodologies that have been successfully applied to other human pathogens [
15‐
17]. In contrast to cross-sectional studies, genetic data from longitudinal historical samples can be used to separately estimate the effective population size (
N
e
) and substitution rate at each locus. In addition, other mutations associated with drug resistance in
P. falciparum were explored to determine how their frequency changed after the drugs were no longer in use.
Discussion
The prevalence of
P. falciparum malaria in Turbo, Colombia fluctuated dramatically between 2003 and 2010 (Fig.
3). These demographic changes were contemporaneous with changes in treatment protocols, which saw the adoption of new drug therapies in 2002, 2007 and 2008. However, as can be seen in Table
1a, the relationship between the treatment protocol and annual malaria prevalence is not simple. For example, whereas the switch to AS + MQ in 2007 had little immediate impact on the number of reported cases, the switch to COARTEM in 2008 coincided with a sharp decline in the number of cases, which decreased from 856 in 2007 to 220 in 2008.
To better understand the processes influencing P. falciparum malaria in this region, this study involved the characterization of this parasite population using a combination of neutral microsatellite markers and loci linked to known drug resistance mutations at three loci. Because these two classes of loci are expected to differ in both their mutation rates and exposure to selection, they have the potential to reveal different features of malaria epidemiology.
The analysis of the microsatellite data revealed the existence of several related multilocus genotypes that persisted for at least 6 years in the region around Turbo (Fig.
1; Additional file
2). This is the pattern expected in a parasite population where control efforts have left behind lineages that became the founders of clusters of highly related parasites [
7,
10]. This observation differs from a previous study of a
Plasmodium population in Peru which found that multilocus genotypes were rapidly broken down by recombination even in low-transmission areas [
31]. Furthermore, all of the multilocus clusters detected in the final year of this study (2009) were also represented in 2002, suggesting that there was little immigration of novel parasite genotypes from outside of the region.
In low-transmission areas, meiotic recombination is most likely to occur between lineages that are closely related, leading to high levels of inbreeding. This is consistent with the fact that many of the putatively multiple infections identified in our sample vary at a single locus, possibly due to mitotic mutations that have occurred within the infected individual. Such mitotic events can lead to an overestimation of the number of observed multiple infections [
8]. Inbreeding can also account for the significant levels of multilocus linkage disequilibrium documented in every year except 2002 (see “
Results”). On the other hand, the microsatellite data suggests that there has been some recombination among the different clusters segregating in Turbo, as is evidenced by the admixture patterns seen in the cluster analysis (Fig.
1).
Although several multilocus haplotypes were observed to persist for multiple years, the cluster analysis revealed pronounced changes in the genetic composition of the
P. falciparum population in Turbo from year to year, with different clusters dominant in different years. This observation is confirmed by the pairwise fixation indices shown in Table
2, which show significant genetic differentiation between every pair of years except between 2009 (when only 14 individuals were sampled) and the preceding 2 years. Rapid non-directional changes in the genetic composition of a population, such as observed here, can be explained by demographic stochasticity (i.e., genetic drift), which is expected to be pronounced in small populations. This is consistent with the low effective population size revealed by the EBSP (Fig.
3).
There is a lack of a correspondence between the estimated
Ne displayed by the EBSP (using the posterior median of the EBSP) and the monthly case reports. During 2003 to 2007,
N
e
remained relatively unchanged even though there were substantial fluctuations in prevalence. Furthermore, between 2007 and 2009 there was a gradual rise in
N
e
even though the number of monthly case reports rapidly decreased following 2008. The reduction in cases coincides with an intense malaria control programme known as “
Papa Luis” carried out in Antioquia between 2007 and 2009. This programme consisted of massive administration of chloroquine (an effective drug to treat
P. vivax but not
P. falciparum), biological control of
Anopheles larvae, swamp drainage and indoor residual spraying [
32]. During this period, the introduction of artemisinin-based combination therapy (ACT) is also thought to have contributed to a reduction in malaria cases.
The discrepancy between the epidemiological and genetic estimates could be explained by several factors. For parasites,
N
e
depends not only on prevalence, but also on the transmission rate and the variance in the number of new cases transmitted by infected individuals [
9,
10,
33,
34]. In particular, a fixed parasite generation time of 1 month was assumed for the study, but the time between transmission events could vary affecting the parasite generation time. Similarly, the number of new cases generated by infected individuals can vary over time due to changes in social dynamics or public health policy and this too will affect
N
e
. The observation that during 2003 to 2007,
N
e
remained relatively unchanged even though there were substantial fluctuations in prevalence suggests that seasonal clonal expansions led to a high prevalence at some time points. Indeed, frequent clonal expansions could even reduce
N
e
if they were accompanied by an increase in either the transmission rate or the transmission variance between infected individuals.
Another factor that may have contributed to the mismatch between the estimated values of Ne and the case report numbers is the variation in the sample sizes collected for genetic analysis in different years. In particular, the apparent rise in the estimated N
e
from 2007 to 2009 may be due to the smaller number of samples collected in these years compared with the period 2003 to 2006. One consequence of the variable sample sizes is that the posterior distribution of the effective population size is more strongly influenced by the prior distribution in those periods when the data are less informative. Indeed, this is evident in the increasing credibility intervals for N
e
between 2007 and 2009. Because of this sensitivity to sample sizes, archiving samples for genotyping should be a standard policy since those will facilitate understanding potential re-introductions or changes in prevalence.
The genetic data from samples collected over a span of 8 years also allowed to estimate microsatellite substitution rates separately from
N
e
. At the six polymorphic loci with enough information, estimates ranged from 5.35 × 10
−3 to 3.77 × 10
−2 substitutions per locus per month (Table
5). In contrast, Su et al. estimated a genome-wide average microsatellite mutation rate of 1.59 × 10
−4 per locus per meiosis from a genetic cross that was typed at 901 loci [
35]. Notably, this rate falls below the 95 % HPD interval of five of the six loci that were investigated. The large discrepancy between these estimates could be due to two differences between the studies. First, whereas Su et al. only considered changes arising during meiosis, the estimates reported here reflect the accumulation of mutations over month-long periods during which the parasite is likely to have undergone multiple mitoses and at most one or two meioses. For the purposes of demographic/epidemiologic inferences, this composite rate is more relevant than the meiotic mutation rate. Secondly, locus-specific variation in microsatellite substitution rates could also explain this discrepancy. Since variable loci were chosen, those may be evolving more rapidly than the genome-wide pool of loci analysed by Su et al.
It should be noted that the mean substitution rates reported in this paper were obtained using the model best supported by our data, the single-phase proportional linear model (PL1). There was no support for any of the two-phase models of microsatellite evolution implemented in BEAST in which alleles may expand or contract by more than one repeat in a single step, although this too could be specific to the loci analysed in this study. The PL1 model assumes that the overall substitution rate is proportional to the repeat length, that each mutation results in the gain or loss of a single repeat, and that the probability of an expansion is a decreasing linear function of the repeat length, so that above a certain threshold, mutations are more likely to result in a contraction rather than in an expansion of the microsatellite [
30]. As such, the rates reported in Table
5 are estimates of the mean substitution rate with respect to the stationary distribution of the microsatellite length under this model. These estimates provide empirical evidence that polymorphic microsatellite loci can be used to make inferences at time scales that are epidemiologically relevant [
36].
Lastly, although drug resistance mutations can decrease in frequency once the corresponding drugs are no longer in use [
37,
38], there are also endemic areas where such mutations have gone to fixation from intermediate frequencies even after drug pressure has been removed [
4,
5,
20]. This study provides evidence of this process. Moreover, the data show that drug resistance mutations were fixed in a parasite population with a low effective population size and then remained at high frequency even after the drug policy was changed. Thus, anti-malarial drugs will remain ineffective unless there is an influx of sensitive parasites from other areas (e.g., Central America). The finding of a sensitive and genetically unrelated parasite demonstrates that such re-introductions are possible. It is worth noting that sensitive
Pfdhps genotypes were still segregating in the population.