Microgeographic Epidemiology of Malaria Parasites in an Irrigated Area of Western Kenya by Deep Amplicon Sequencing

Abstract

To improve food security, investments in irrigated agriculture are anticipated to increase throughout Africa. However, the extent to which environmental changes from water resource development will impact malaria epidemiology remains unclear. This study was designed to compare the sensitivity of molecular markers used in deep amplicon sequencing for evaluating malaria transmission intensities and to assess malaria transmission intensity at various proximities to an irrigation scheme. Compared to ama1, csp, and msp1 amplicons, cpmp required the smallest sample size to detect differences in infection complexity between transmission risk zones. Transmission intensity was highest within 5 km of the irrigation scheme by polymerase chain reaction positivity rate, infection complexity, and linkage disequilibrium. The irrigated area provided a source of parasite infections for the surrounding 2- to 10-km area. This study highlights the suitability of the cpmp amplicon as a measure for transmission intensities and the impact of irrigation on microgeographic epidemiology of malaria parasites.

This study highlights the usefulness of targeted deep sequencing of the cpmp amplicon in revealing malaria transmission patterns at a microgeographic spatial scale (<20 km diameter) and the impacts that irrigated agriculture has on the epidemiology of malaria parasites.

Development of irrigation schemes has been linked to an increase in malaria risk by providing more habitats for mosquito breeding, enhancing habitat stability, and/or providing vectors to bridge transmission seasons [1]. For example, irrigated rice fields in a subarid region of Madagascar were found to have a transmission rate 150 times higher than in the original ecosystem; additionally, the region transformed from a seasonal transmission area to a perennial transmission area [2]. However, there are examples where there has been no observed impact or even a reduction in malaria prevalence following water resource development projects [3]. For instance, the Diama Dam in Senegal had no impact on malaria transmission despite an increase in mosquito vector abundance [4, 5]. In contrast, the construction of small dams in the Sundergarh district of India led to a decrease in local malaria prevalence [6], which was attributed to increased flow conditions that rendered mosquito vector breeding unfavorable in the river downstream of the dams. Thus, the impacts that water resource development projects have on malaria risk are likely context specific, depending on the local epidemiologic setting. Malaria epidemiological studies can be enhanced by leveraging data from the genetic diversity of parasites. Individuals in malaria-endemic areas often harbor multiple, genetically distinct strains, known as polyclonal infections. Polyclonal infections can result from 2 processes: (1) superinfection, which occurs from subsequent bites by multiple, infected mosquitoes, each carrying a unique parasite genotype; or (2) co-transmission, when an individual is bitten by a single mosquito, in which the mosquito is carrying multiple parasite genotypes [7]. Furthermore, when multiple parasite genotypes are present in a mosquito host during the sexual stage, a high rate of genetic recombination occurs [8], generating novel local genotypes [7, 9] and thus leading to more opportunities for polyclonal infections to occur. Therefore, the number of parasite strains coinfecting a single host, called the multiplicity of infection (MOI), is expected to positively correlate with the intensity of malaria transmission because of increased opportunities for superinfection and co-transmission to occur in high-transmission areas [7, 10].

While MOI may be a useful indicator for transmission intensity, in some studies, there has been no observed positive correlation between infection complexity and malaria parasite prevalence [11, 12]. However, inconclusive correlations between MOI and transmission intensity may be in part due to the choice of genotyping method and/or molecular marker(s), which can affect the ability to detect differences in MOI at varying transmission intensities. Polymerase chain reaction (PCR)–based genotyping, such as by size-polymorphic antigens or microsatellites, often underestimates MOI, as it is challenging to detect minority strains, which result in faint bands or weak fluorescent signals [13]. In contrast, deep amplicon sequencing has been shown to be more sensitive for detecting MOI [14–16]. Moreover, choice of molecular marker for amplicon sequencing, which can vary in terms of the number of single-nucleotide polymorphisms (SNPs), can also impact the ability to assess MOI [14, 16] and, thus, potentially the sensitivity to assess differences in MOI between varying transmission intensities. Recently, Lerch et al [17] demonstrated that cpmp (PF3D7_0104100, “conserved Plasmodium membrane protein”) detected a higher mean MOI and contained more SNPs than common molecular markers within parasites originating from Papua New Guinea.

Therefore, the objectives of this study were the following: (1) examine the level of sequence polymorphism found in cpmp, ama1, csp, and msp1 markers in Plasmodium falciparum by deep amplicon sequencing; (2) evaluate P. falciparum malaria transmission intensity at various proximities to an irrigation scheme by PCR positivity rate, MOI, and other molecular indices; (3) compare the sensitivity of molecular markers in detecting differences in MOI among the various transmission intensity zones; and (4) assess patterns of genetic connectivity among parasite isolates among irrigated transmission risk zones.

MATERIALS AND METHODS

Ethics Statement

Scientific and ethical clearance was obtained from the institutional scientific ethical review board of the University of California, Irvine and Maseno University, Kenya. Written informed consent/assent for study participation was obtained from all consenting heads of households, parents/guardians (for minors <18 years of age), and each individual who was willing to participate in the study.

Study Design

This study was conducted in the Oluch Irrigation Scheme (latitude 0°26′44′′S, longitude 34°31′28.0′′E, elevation 1200 m) and its vicinity in Homa Bay County, western Kenya (Figure 1). This area falls within the lake-endemic zone, which has year-round malaria transmission with seasonal variability [18]. To build sustainable agriculture and reduce poverty, the Kimira-Oluch Smallholder Farm Improvement Project was funded by the African Development Bank. As part of this project, the Oluch irrigation scheme was initiated in 2007 and completed in 2015. Clusters from the Oluch irrigated scheme and surrounding areas were assigned to the following classes: high (ie, within the irrigated area), medium (ie, within a 2-km area from the boundary of the irrigation scheme), and low (ie, >5 km from the boundary of the irrigation scheme). Cluster radii varied from 0.25 to 1.0 km.

Figure 1.

Study site localities in relation to the Oluch irrigation scheme, Kenya. Abbreviations: Hi, high; Lo, low; Me, medium.

Sample Collection

Fingerprick blood samples were collected during 2 cross-sectional surveys in February–March and June–July 2018. The February–March collection period occurs after the short rains, whereas June–July occurs after the long rains. A random subset of residents, ranging from 19 to 86 per cluster, was selected for participation in this study. Study participants reported having no malaria symptoms at the time of collection, so that only parasites from subclinical infections were examined. Whatman 3MM filter papers were used for collecting and storing approximately 50 µL of blood. DNA was extracted and purified from dried blood spots by the Saponin/Chelex method [19]. A multiplexed TaqMan probe assay was performed to identify malaria parasite species [20, 21]. Of those, 91 samples that were infected with the P. falciparum parasite species only were randomly selected for evaluation by deep amplicon sequencing.

Amplicon Deep Sequencing

Amplification and sequencing of the molecular markers ama1, cpmp, csp, and msp1 were performed based on previously published primers and protocols [14, 16, 17, 22, 23] with modifications. In brief, the sequencing library was generated by 3 rounds of PCR. The primary PCR amplified each relevant molecular marker. The second round was a nested, marker-specific amplification primers that carried a 5′ linker sequence. PCR products of each molecular marker were then pooled by sample, and the third round of PCR was completed using primers with sample-specific barcode sequences to allow for pooling of samples and subsequent de-multiplexing. Plasmodium falciparum laboratory strains HB3 (MRA-155G), DD2 (MRA-150G), and 3D7 (MRA-102G) were used as controls in duplicate at the ratios 1:0:0, 7:3:0, and 6:3:1, respectively. Thirty samples were amplified in duplicate to assess potential PCR or sequencing errors. Amplicons were cleaned and normalized to 1 ng/µL concentration using the SequalPrep Normalization Plate Kit (ThermoFisher Scientific, Waltham, Massachusetts). Sequencing was performed on an Illumina Miseq using a MiSeq reagent kit v3 PE300 (UCI Genomics High-Throughput Facility) with PhiX control (Illumina, PhiXControl v3). PCR protocols are described further in the Supplementary Materials, including primer sequences (Supplementary Table 1), PCR reaction mixtures (Supplementary Table 2), and thermocycling conditions (Supplementary Table 3).

Bioinformatic Analysis

Paired ends were joined with fastq-join (parameters: 8% maximum percentage difference and 20 minimum overlap). Pooled reads were de-multiplexed by molecular marker with seqkit grep by matching 10 base pairs at the 3′ end of the forward marker-specific primers. Adapter and primer sequences, as well as low-quality base pairs (parameter: error threshold 0.01) were subsequently trimmed by seqtk trimfq. Additionally, de-multiplexed samples by molecular marker yielding <25 reads’ sequencing coverage were excluded. Haplotypes were determined with the use of SeekDeep software [24]. Specifically, SeekDeep qluster was run (parameter: illumina) to create haplotypes with relative abundances by collapsing reads on specific errors. Next, for replicate comparison and final results filtering, SeekDeep processClusters was run (parameters: strictErrors, illumina, fracCutOff 0.035, clusterCutOff 3, and hq 1), in which the parameters allowed for a few low-quality mismatches but no indels and 1 high-quality mismatch, cleared out low abundant clusters (3.5% and lower), and removed clusters of size 3 or less. These parameters resulted in consistent and accurate haplotype calling among positive controls and sample replicates for all molecular markers.

Data Analysis

All data analyses were performed in R software unless otherwise noted. The statistical significance of differences in PCR positivity rate among risk zones was assessed by χ ² test with a Bonferroni correction to limit the type 1 error rate. The 95% confidence intervals were estimated by the modified Wald method. The significance of MOI differences among risk zones was evaluated by Wilcoxon rank-sum tests with a Bonferroni correction due to the nonnormal distribution of MOI values. Sample size simulations were carried out by drawing from a Poisson distribution [25, 26], with lambda (λ) equal to the average MOI by molecular marker and risk zone. Each simulation was performed with 10 000 replicates. Nucleotide diversity and linkage disequilibrium (LD) were assessed for cpmp amplicons in DNA Sequence Polymorphism v5 by coalescent simulations (parameters: free recombination and 1000 replicates) [27]. Differences in nucleotide diversity and LD were assessed by the pairwise comparison method assuming normal distributions. Proportion of shared alleles was measured by cpmp amplicons using the R package adegenet [28]. Isolates that shared >98% of polymorphic alleles (≤1 nucleotide difference), henceforth referred to as highly related pairs, were visualized and evaluated for patterns of allele sharing. Differences in proportions of highly related pairs among risk zones were evaluated by a χ ² test with a false discovery rate correction. Differences in geographic distances between highly related isolates and all isolates were assessed by fit of analysis of variance followed by TukeyHSD post hoc analysis. Finally, to assess the magnitude and directionality of parasite migration among risk zones, a Bayesian inference method based on coalescent theory was implemented in Migrate-N 3.7.2 [29, 30]. Parameters estimated from genetic data include mutation-scaled immigration rate (M), mutation-scaled population size (θ), and the number of effective migrants per generation (Nm, calculated as θM/2). Ten independent runs were conducted with a burn-in of 10⁶ steps, sampling increment of 10 steps, and 10⁶ recorded steps in each chain for a total of 10⁷ visited parameter values.

RESULTS

PCR Positivity Rate Among Transmission Risk Zones

In total, 2112 samples were evaluated for the presence of Plasmodium malaria parasites from 21 study sites within and surrounding the Oluch irrigation scheme (Figure 1). In February–March, the PCR positivity rate was significantly higher in the high-risk zone (21.0% [74/353]), as compared to the medium-risk (11.1% [45/407]) and low-risk (11.0% [32/290]) zones (χ ² = 13.31, P = .0006 and χ ² = 10.69, P = .002, respectively) (Figure 2). In June–July, the PCR positivity rate was highest in the medium zone (18.5% [65/352]) followed by high (16.2% [81/500]) and low (14.8% [31/210]), but pairwise differences were not significant (χ ² ≤ 1.03, P ≥ .93 for all comparisons) (Figure 2).

Figure 2.

Polymerase chain reaction (PCR) positivity rate among transmission risk zones and season, Kenya. Square indicates overall positivity rate among clusters by risk zone. Lines indicate 95% confidence intervals. Asterisks indicate statistical significance (χ ² test with Bonferroni correction, P ≤ .05). February–March occurs after the short rains (October–December), and June–July occurs after the long rains (March–May).

Sequence Reads and Haplotype Determination

A random subset of P. falciparum–infected samples were selected for evaluation by deep amplicon sequencing. A total of 116 pooled PCR reactions (85 samples, 25 replicate PCR reactions, and 6 controls) were successfully amplified, sequenced, and joined, resulting in 515 952 reads. Average depth of reads per sample was 4634 (range, 206–9560). Of the 4 molecular markers, sequencing by cpmp detected the highest number of unique haplotypes (78 haplotypes), the highest percentage of polymorphic infections (49.1%), and highest average MOI (2.02) (Table 1 and Supplementary Figure 1). These trends were consistent when comparing values among all samples and matched samples only (Supplementary Table 4). Average MOI was generally highest in the medium-risk zone and lowest in the low-risk zone across molecular markers (Table 1).

Table 1.

Comparison of Infection Complexity Among Molecular Markers and Transmission Risk Zones, Kenya

Marker	Sample Size, No.	No. of Haplotypesa	% Poly.b	Average MOIc
				High	Medium	Low	Overall
ama1	46	20	34.8	1.41	1.63	1.60	1.52
cpmp	55	78	49.1	2.08	2.47	1.17	2.02
csp	56	39	41.6	1.61	2.00	1.44	1.71
msp1	77	55	42.9	1.91	2.05	1.15	1.79
Comb.d	85	…	55.3	2.24	2.55	1.55	2.19

Abbreviation: MOI, multiplicity of infection.

^aTotal number of unique haplotypes detected.

^bPercentage of samples that had >1 haplotype (polyclonal).

^cAverage MOI among samples.

^dPooled values for all molecular markers; ie, the maximum MOI per sample is obtained of the 4 molecular markers.

Sample Size Simulations

To assess the power of discriminating between transmission intensities by MOI estimated from sequencing of the 4 molecular markers, sample sizes (up to 5000) were simulated for each molecular marker and compared. Only by sequencing of cpmp or msp1 was an average unadjusted P ≤ .0167 achieved among 10 000 simulations when comparing high- vs low-risk and medium- vs low-risk zones at sample size ≤66 per risk zone (Figure 3A). The smallest sample size required to obtain significant differences between the high- and low-risk zone in ≥80% of 10 000 simulations was lowest for cpmp (45 at α = .0167; 33 at α = .05) (Figure 3B). Likewise, sequencing by cpmp required the lowest sample size to achieve significant differences between the medium- and low-risk zone (25 at α = .0167; 19 at α = .05) (Figure 3B). Detecting significant differences between the high- and medium-risk zones required sample sizes exceeding 150 for all molecular markers, but was lowest for csp (261 at α = .0167; 196 at α = .05) (Figure 3B). Last, to evaluate how simulated P values compared to empirical P values from this study, we visualized the distribution of P values derived from 10 000 simulations at sample sizes equivalent to the empirical sample sizes (Figure 3C). Eight of the 12 empirical P values were within the middle 50% of P values from respective simulations (Figure 3C).

Figure 3.

Simulated sample sizes to assess power of hypothesis testing by molecular markers, Kenya. A, Mean unadjusted P value by simulated sample size. Dashed red line indicates α = .167. Dotted red line indicates α = .05. B, Minimum sample size for unadjusted P value to be less than α in ≥80% of simulations. C, Empirical P value compared to distribution of simulated P values at equivalent sample sizes. Diamonds indicate empirical values. Box plots indicate distribution of simulated values. Lower and upper hinges correspond to the first and third quartiles. All results are based on Wilcoxon rank-sum tests among 10 000 simulations per sample size. Sample sizes were simulated from Poisson distributions with λ = empirical average multiplicity of infection for a given molecular marker and transmission zone.

Comparison of Genetic Indices Among Transmission Risk Zones by cpmp

Since amplicon sequencing by cpmp detected the highest average MOI and required the lowest sample sizes to detect significant differences in MOI in 2 of 3 comparisons by simulated data, subsequent analyses were based on cpmp amplicons exclusively. Average MOI was significantly higher in the medium-risk zone than in the low-risk zone (P = .004; Wilcoxon rank-sum test) (Figure 4A). MOI did not vary significantly between the high- and medium-risk zone or between the high- and low-risk zone (P = .74 and P = .06, respectively). Notably, MOI did not vary significantly among other risk factors, including sex (P = 1), age group (P ≥ .27 for all comparisons), and season of collection (P = 1) (Supplementary Figure 2). Average nucleotide diversity did not differ significantly among transmission risk zones based on confidence intervals obtained from coalescent simulations (P > .05) (Figure 4B). LD was significantly lower in the high- and medium-risk zones (0.020 and 0.022, respectively) as compared to the low-risk zone (0.077) (P = .05; Figure 4C).

Figure 4.

Genetic indices among transmission risk zones by cpmp amplicon sequencing. A, Multiplicity of infection among risk zones. Dots indicate individual data points. Triangles indicate average values. B, Nucleotide diversity among risk zones. C, Linkage disequilibrium among risk zones. Squares indicate average values among 1000 coalescent simulations. Lines indicate 95% confidence interval from simulations. Asterisks indicate statistical significance.

Allele Sharing and Patterns of Relatedness Among Parasite Isolates by cpmp

To assess patterns of genetic connectivity, highly related isolate pairs (sharing >98% of alleles) were identified and mapped by geographic locality (Figure 5A). Four pairs of highly related isolates originated from the same cluster, and so are not visible on the map: high (1 pair), low (2 pairs), and medium (1 pair). The most common allele sharing pattern was between isolates originating from the high- and medium-risk zones (Figure 5B). Moreover, geographic distance alone did not explain this allele sharing pattern (Figure 5C), as geographic distance did not significantly differ among highly related pairs, identical pairs, and all pairs (P ≥ .13 for all comparisons; unpaired t test with Bonferroni correction). Finally, an analysis of migration rates revealed that the highest magnitude of migration among risk zones occurred in the direction of the high- to the medium-risk zone (Nm = 25.4), which was followed by medium- to low-risk zone (Nm = 18.8), and then high- to low-risk zone (Nm = 17.4), indicating that areas closer to the irrigation scheme provide sources of parasites for surrounding areas (Figure 6).

Figure 5.

Allele sharing among parasite isolates by cpmp amplicon sequencing, Kenya. A, Highly related isolates by geographic locality. Dots indicate individual parasite isolates. Lines connect isolates from different clusters that share >98% of polymorphic alleles. Thin lines indicate 98%–99.9% allele sharing; thick lines indicate 100% allele sharing. B, Highly related pairs by risk zones. Bar height indicates the proportion of highly related pairs among all isolate pairs for each risk zone combination. Labels indicate the total count number of highly related pairs for each risk zone combination. For x-axis labels, letters indicate risk zone combinations (H, high; L, low; M, medium). Asterisks indicate statistical significance (χ ² test with false discovery rate correction, P ≤ .05). Four pairs of highly related isolates originated from the same cluster, and so are not visible on the map: high/high (1 pair), low/low (2 pairs), and medium/medium (1 pair). C, Histogram of highly related pairs by geographic distance between pairs. Solid green line indicates density plot. Vertical dashed line indicates mean geographic distance between isolate pairs. “All” indicates geographic distances between all isolate pairs. Geographic distances were not significantly different between allele sharing categories (unpaired t test with Bonferroni correction, P ≥ .13 for all comparisons).

Figure 6.

Migration rates among risk zones by cpmp sequencing, Kenya. Thickness of arrows is proportional to the estimated mutation-scaled migration rate (M). Diameter of circles is proportional to the estimated mutation-scaled population sizes (θ). Numbers indicate the estimated effective number of migrants per generation (Nm, calculated as θM/2) between populations. Black arrows indicate predominant direction of migration.

DISCUSSION

This study examined the utility of 4 molecular markers (ama1, cpmp, csp, and msp1) used for deep amplicon sequencing in evaluating malaria transmission intensities and assessed malaria transmission intensity and genetic connectivity among various proximities to an irrigation scheme. We found that amplicon sequencing by cpmp had the highest sensitivity to detect transmission intensity differences based on MOI. Additionally, we found that indicators of transmission intensity were highest within 5 km of the irrigation scheme, as compared to 5–10 km from the irrigation scheme. This finding is in agreement with previous studies in Africa that documented significantly higher malaria intensity at close proximity (<5 km) to irrigation dams than those located farther away (>5 km) [31–34]. Furthermore, based on cpmp amplicon data, we demonstrated that the area within 2 km of the irrigation scheme provides a source of parasite infections for the surrounding 2- to 10-km area. These findings highlight the value of the cpmp amplicon in studying microgeographic epidemiology of malaria and that irrigated agriculture promotes a source of parasite infections for surrounding areas in this epidemiologic setting.

Quantifying malaria transmission intensity can be done through several indices, such as the traditional metrics of entomological inoculation rate (EIR) or clinical incidence. However, measuring EIR is particularly labor-intensive [35], and neither of these metrics account for polymorphic infections. Thus, genotyping-based metrics provide an appealing alternative or supplement to quantify malaria transmission intensity. In particular, MOI is an attractive indicator for assessing transmission intensity, as it requires a modest sample size [36] and can be measured from a single time point [26]. Consistent with a study in Papua New Guinea [17], we found that the cpmp amplicon detected the highest average MOI among common molecular markers by deep amplicon sequencing. Furthermore, we demonstrated that this enhanced sensitivity to detect MOI resulted in requiring a smaller sample size than alternative markers to observe differences among transmission risk zones in this study. Therefore, MOI assessed by cpmp amplicon sequencing is a relatively low-cost metric, requiring a rather low sampling effort.

Additionally, we showed that deep sequencing by cpmp can be useful for revealing source-sink parasite dynamics [29, 30]. Identifying reservoirs or factors that promote reservoirs of malaria parasites can be used to plan effective interventions for malaria control and elimination [36]. Traditional molecular markers, such as microsatellites or SNP barcodes, are also useful for inferring patterns of genetic connectivity [37–39]. However, constructing haplotypes from these multilocus markers is not feasible for polymorphic infections. As a result, in areas of high transmission, where polymorphic infections are predominant, a large amount of data is lost or thrown out when using common multilocus markers. Thus, deep amplicon sequencing provides a solution to tracking malaria parasites and identifying source-sink patterns without losing information from polymorphic infections.

Across sub-Saharan Africa, irrigation has been blamed for intensifying malaria in areas with unstable disease transmission [31, 40]. There is evidence demonstrating that irrigation schemes create conducive microclimates for malaria mosquitoes to breed [41, 42] and promote longevity in adult mosquitoes [43]. Behaviorally, people living close to irrigated fields spend more time outside on their field during the early hours of the night where they are not protected by indoor-based control measures (eg, bed nets, indoor residual spraying) while the mosquitoes are active [44, 45]. Increased malaria transmission intensity coupled with outdoor behavior could potentially compromise the efficacy of existing malaria intervention efforts in the irrigated areas that mainly rely on indoor-based control tools. As Africa is planning to achieve malaria elimination in many of its regions by 2030 [46], such localized transmission pockets could challenge intervention efforts in the region. Thus, additional vector control strategies are critically needed to address outdoor transmission. Integrated vector management by incorporating larval management through irrigation canal management may help reduce malaria around irrigated settings, as suggested in previous field studies [47, 48].

This study had certain limitations. PCR positivity rate and MOI, like all indicators of malaria transmission, have inherent biases. For example, parasite positivity rate is affected by acquired immunity and antimalarial drug use, and it does not capture multiple infections [26]. MOI is also mediated by acquired immunity and antimalarial drug use. Additionally, MOI is limited by the diversity of parasite populations—that is, where parasite populations are less diverse, MOI may be an underestimate of transmission because multiple clones are indistinguishable [49]. The ability to detect MOI is limited by the selected limit of detection, which was conservatively set at 3.5% for this study. As a result, it is likely we missed minority genotypes below this threshold, but this trade-off is balanced by filtering out potential noise from PCR and sequencing errors [13]. We found that decreasing this threshold increased variability among replicates. Lowering the limit of detection may be possible by decreasing the number of PCR cycles and increasing the amount of starting genetic material instead to limit potential PCR errors. With that said, the overall trends observed in this study should not be affected by these limitations.

To conclude, we demonstrated the importance of selecting a highly polymorphic molecular marker used for amplicon sequencing for detecting differences in malaria transmission intensities. We found that deep sequencing of cpmp required the smallest sample size of the 4 molecular markers to detect significant differences in MOI among transmission risk zones. Transmission intensity by all indices was lowest 5–10 km from the irrigation scheme compared to <5 km from the irrigation scheme. Moreover, the high migration rates from the irrigated area to surrounding areas suggests that the Oluch irrigation scheme is promoting a source of parasite infections for nearby areas. These findings highlight the usefulness of sequencing by cpmp in detecting patterns of malaria transmission at a microgeographic spatial scale (<20-km study area), as well as the impacts that irrigated agriculture has on microgeographic epidemiology of malaria parasites.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We thank the study participants and the field team of the International Centers of Excellence for Malaria Research (ICEMR) program in Kenya for their involvement in this study. We thank Nicole Hemming-Schroeder for her contribution in coding the sample size simulations. The sequences for this project have been deposited at GenBank under accession numbers MT818243–MT818353.

Financial support. This work was supported by the National Institutes of Health (grant numbers U19AI129326, R01AI050243, D43TW001505, and F32AI147460).

Potential conflicts of interest. All authors: No reported conflicts of interest.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.