Monitoring the changing spatial distribution of drug resistance markers is necessary
for developing efficient interventions to halt the spread of resistance and eliminate
malaria. Here, we leverage geolocated samples from the DHS to measure resistance mutations
across the DRC and map changes that occurred between 2007 and 2013 (6,19,21). Studies using nationally representative samples like the DHS are less susceptible
to selection bias; however, longitudinal comparisons of DHS surveys have been impeded
by the fact that the individual survey clusters change between surveys. Here, we use
a spatial prediction model that overcomes this by assuming a continuous surface of
underlying allele frequencies, allowing us to integrate information at different points
in space and time.
This study found that the 540 and 581 pfdhps mutations have increased in the DRC since 2007, both in numbers and in geographic
spread. This agrees with recent findings of an increase in pfdhps mutations between 2014 and 2015 amongst individuals living in southwest DRC (45). Evidence of geographic expansion from the eastern part of DRC is also supported
by previous research that demonstrated higher prevalence of both mutations in East
Africa compared to West Africa (9,46). This expansion is particularly concerning as these mutations are associated with
SP failure during IPTp (11,19,46). The risk factor analysis indicates that these increases may be in part driven by
SP use, which was associated with increased prevalence of pfdhps mutations. Further, this study indicates that increasing community level drug use,
not necessarily individual use, is associated with increases in resistance. This is
consistent with previous work that demonstrated associations between community level
interventions and malaria risk (34,35).
Chloroquine resistance has remained relatively steady since 2007; the proportion of
CVIET parasites is unchanged and the spatial distribution remains similar. These findings
are troubling as the DRC halted chloroquine use as a first line treatment in 2001
due to concerns about growing resistance (47,48). This sustained resistance may be in part driven by demographic factors; the risk
factor models results indicate that cluster-level education and wealth are associated
with chloroquine resistance. There may also be unregulated chloroquine use, as has
been reported in other sub-Saharan African countries (49). Additionally, there is evidence that the CVIET haplotype is associated with amodiaquine
resistance (15,16,50). Since amodiaquine is used as part of the first line treatment ASAQ in the DRC, this
association may explain why the prevalence of CVIET has remained steady over time
(16,48,50). Reported ASAQ use was too low in this study for us to evaluate this relationship
statistically. However, we did not detect the SVMNT haplotype, also found to be associated
with ASAQ resistance, in this population (17,18).
The findings from this study have direct implications for malaria control programs
in the DRC. As mentioned, SP is still used in the DRC as the primary drug for IPTp
(47,48). Increasing SP resistance may threaten these preventive efforts. Additionally, though
chloroquine is no longer a recommended treatment for malaria, reports from other sub-Saharan
African countries show a steep drop in the proportion of resistant parasites after
ending chloroquine use (48,51–53). The sustained prevalence of chloroquine resistance seen in this study is alarming
and warrants further investigation.
Effective monitoring of drug resistance requires sensitive molecular tools that can
accommodate a large number of samples. Using MIPs to amplify resistance loci allows
for highly multiplexed and efficient deep sequencing of Plasmodia. This study demonstrates the utility of MIPs for drug resistance surveillance, and
the ability to answer critical epidemiological questions. This novel method can also
be used to investigate questions of parasite population structure, gene flow, and
selective sweeps, amongst others. The spatial-temporal approach used here also represents
a step forward compared with previous mapping efforts (19). The random Fourier features (RFF) method allows us to explore complex models in
a computationally efficient way, thereby reducing the time and resources required
to perform this kind of advanced spatial analysis and opening the door to much larger
datasets in the future.
There are several limitations to this study. First, we only have access to a relatively
small number of samples distributed over a wide geographic area, and this is reflected
in the large credible intervals around our spatial-temporal predictions. We can therefore
only draw large-scale conclusions about changes that have occurred over the study
time period, based on patterns that are consistent over the majority of posterior
draws. Second, this study compared genotype data generated using different approaches:
data from 2013-2014 was obtained using MIPs and Illumina sequencing, while data from
2007 was obtained with standard PCR amplification and alternate sequencing methods.
However, the sequencing coverage is approximately the same across studies, providing
assurance that the methods are comparable. Additionally, the MIPs did not amplify
across all of pfdhps in a single sequence but rather used multiple MIP probes to target the regions of
interest. Therefore, we could not create true haplotypes across pfdhps.