The impact of indoor residual spraying on Plasmodium falciparum microsatellite variation in an area of high seasonal malaria transmission in Ghana, West Africa

Abstract Here, we report the first population genetic study to examine the impact of indoor residual spraying (IRS) on Plasmodium falciparum in humans. This study was conducted in an area of high seasonal malaria transmission in Bongo District, Ghana. IRS was implemented during the dry season (November–May) in three consecutive years between 2013 and 2015 to reduce transmission and attempt to bottleneck the parasite population in humans towards lower diversity with greater linkage disequilibrium. The study was done against a background of widespread use of long‐lasting insecticidal nets, typical for contemporary malaria control in West Africa. Microsatellite genotyping with 10 loci was used to construct 392 P. falciparum multilocus infection haplotypes collected from two age‐stratified cross‐sectional surveys at the end of the wet seasons pre‐ and post‐IRS. Three‐rounds of IRS, under operational conditions, led to a >90% reduction in transmission intensity and a 35.7% reduction in the P. falciparum prevalence (p < .001). Despite these declines, population genetic analysis of the infection haplotypes revealed no dramatic changes with only a slight, but significant increase in genetic diversity (H e: pre‐IRS = 0.79 vs. post‐IRS = 0.81, p = .048). Reduced relatedness of the parasite population (p < .001) was observed post‐IRS, probably due to decreased opportunities for outcrossing. Spatiotemporal genetic differentiation between the pre‐ and post‐IRS surveys (D = 0.0329 [95% CI: 0.0209 – 0.0473], p = .034) was identified. These data provide a genetic explanation for the resilience of P. falciparum to short‐term IRS programmes in high‐transmission settings in sub‐Saharan Africa.


| INTRODUC TI ON
Indoor residual spraying (IRS) is a widely used public health intervention to reduce and ultimately interrupt malaria transmission by decreasing the lifespan of the female Anopheles mosquito (i.e., vector) (Bhatt et al., 2015;Pluess et al., 2010;World Health Organization, 2014, 2015b. This impact on transmission also leads to reduced incidence and prevalence of Plasmodium spp. infection in humans. To date there have not been any studies on the impact of IRS on parasite population genetics, yet it is reasonable to propose that reduced transmission and infection prevalence would lead to decreased Plasmodium falciparum diversity within individual human hosts and overall, in the population. Considering that sexual recombination (i.e., meiosis) is an obligatory part of the P. falciparum life cycle in the mosquito (Paul & Day, 1998), IRS should also lead to reduced outcrossing in the mosquito with consequences for the genetic structure of the parasite population.
These studies have shown that in high-transmission settings (e.g., sub-Saharan Africa) where the majority of infections are multiclonal, the P. falciparum population is characterised by high diversity, low levels of population differentiation, and linkage equilibrium, increasing the likelihood of recombination between genetically distinct parasite clones (i.e., outcrossing) in the mosquito following a blood meal. In contrast, low genetic diversity, extensive population differentiation, and strong linkage disequilibrium, are typically seen in low-transmission regions (e.g., South America) or areas under intense control suggesting higher levels of inbreeding in these populations.
The latter are "ideal elimination units" that may lead to a population bottleneck and where clonal parasites may be more readily lost to genetic drift events (Cotton et al., 2018;Escalante et al., 2015).
Gene flow from higher transmission neighbouring regions can also increase diversity in areas of lower transmission or under malaria control (Branch et al., 2011;Roh et al., 2019). The high-transmission genetic profile is characteristic of much of sub-Saharan Africa that has been recently prioritized by the Roll Back Malaria Partnership and World Health Organization's (WHO) "High Burden to High Impact" (HBHI) country-led approach to accelerate progress against malaria.
Importantly the HBHI initiative points to IRS as one of the key interventions to enable high-burden countries to get back on track towards malaria elimination (World Health Organization, 2019. Malaria transmission remains highly seasonal across large parts of sub-Saharan Africa, with peaks during the rainy season (i.e., high-transmission season) and troughs in the dry season (i.e., lowtransmission season). Since mosquitoes require pools of water to breed, vector densities along with malaria transmission will increase during the wet season, while during the dry season fewer mosquitoes will propagate, leading to reduced transmission intensity.
Previous studies have shown that in these areas characterized by high seasonal malaria transmission a large proportion of the human population across all ages harbour asymptomatic P. falciparum infections during both the wet and dry seasons (Galatas et al., 2016;Koram et al., 2003;Lindblade et al., 2013;Owusu-Agyei et al., 2002;Tiedje et al., 2017). This reservoir of asymptomatic infections at the end of the dry season initiates transmission at the start of the next wet season and as such presents a target for interventions to potentially reduce or bottleneck the parasite population by vector control with IRS. The impact of this strategy was investigated by measuring microsatellite variation in the P. falciparum population in humans in an area characterized by high seasonal malaria transmission in Ghana, West Africa. 089/11-12), The University of Melbourne, Australia (HREC 144-1986 and HREC 195-5652), and the University of Chicago, United States (IRB 14-1495). Individual informed consent was obtained in the local language from each enrolled participant by signature/thumbprint, accompanied by the signature of an independent witness. For children <18 years of age a parent or guardian provided consent. In addition, all children between the ages of 12 and 17 years provided assent.

| Study area and design
This study to investigate the impacts of an IRS intervention, under operational conditions, on the asymptomatic Plasmodium falciparum reservoir was conducted in Bongo District, located in the Upper East Region of Ghana (Figure 1a). Bongo District has a short but intense rainy season (~70 days with rain per year between June to October) and a prolonged dry season (November to May). Malaria in Bongo District is hyperendemic (i.e., consistent at high levels) and is characterised by marked seasonal transmission of P. falciparum (minor parasites: P. malariae and P. ovale) . For this study, participants were enrolled from two broad "catchment areas" (Vea/ Gowrie and Soe) and hereinafter referred to collectively as "Bongo" for discussion purposes (Figure 1a). The catchment areas were considered to be different agroecological zones (irrigated vs. nonirrigated) based on their proximity to the Vea Dam, but were otherwise similar with respects to population size, age structure, and ethnic composition .
Using a cross-sectional study-design, two age-stratified surveys of ~2000 participants per survey were undertaken at the end of the wet seasons (i.e., high-transmission season) pre-IRS (T1, October 2012) and post-IRS (T2, October 2015) (Figure 1b). During each survey, relevant information was collected from all participants using structured questionnaires on their demographics, malaria history, and malaria prevention activities. Additional details on the study area, study population, sample collection procedures, etc. have been previously published . All individuals who were microscopically positive for P. falciparum (including mixed P. falciparum/P. malariae infections), were afebrile (axillary temperature <37.5°C) on the day the survey was conducted, and did not report a history of fever in the 24 h prior to being surveyed were defined as having an "asymptomatic P. falciparum infection" (hereafter designated as P. falciparum infections).
Over the last decade IRS has been scaled up across northern Ghana through ongoing support from the United States President's Malaria Initiative (PMI) and the Global Fund in partnership with the AngloGold Ashanti Malaria Control Programme (AGAMal) (Figure 1c) (National Malaria Control Programme, 2009The Global Fund, 2012 (Smith Paintain et al., 2014;UNICEF, 2012;USAID Global Health Supply Chain Program, 2020). LLIN usage pre-and post-IRS was high in Bongo, with 89.1% and 90.6% of participants, respectively, reporting sleeping under an LLIN the previous night (Table S1).

| Microsatellite genotyping
For all participants with microscopically confirmed P. falciparum infections (i.e., isolates), two 5 x 5 mm sections were cut from each dried blood spot and placed in a 1.5-ml centrifuge tube, with genomic DNA (gDNA) being extracted using the QIAmp DNA mini kit (Qiagen) as previously described . A subset of 200 microscopic P. falciparum isolates from both the pre-IRS (T1) and post-IRS (T2) surveys were selected for microsatellite genotyping based on their multiplicity of infection (MOI) (i.e., number of genetically distinct P. falciparum genomes) as determined using var genotyping (see Supporting Information Methods, Figure S1). Briefly, using this approach we estimated the MOI based on the number of var DBLα types identified per isolate, using a cutoff value of 60 var DBLα types per P. falciparum genome. Isolates with ≤60 var DBLα types were classified as single-clone infections (MOI = 1), while isolates with >60 var DBLα types were classified as multiple-clone infections (MOI > 1). To facilitate a more accurate assignment of the fluorescent peaks during the analysis (described below), only those isolates with a MOI = 1 or 2, were selected for the microsatellite genotyping (Anderson et al., 1999).
The P. falciparum isolates (N = 400) selected from the preand post-IRS surveys were genotyped using a verified panel of 12 putatively neutral microsatellite markers located across the 14 chromosomes as described by Anderson et al. (1999 ): TA1,  2490, TA81, TA87, TA109, TA60, POLYA, TA42, ARA2, PfG377, PfPK2, and TA40, with modified cycling conditions as specified in Ruybal-Pesántez et al. (2017). Fluorescently-labelled PCR products were sent to a commercial sequencing facility (Macrogen Inc., South Korea) for capillary electrophoresis and fragment analysis on an Applied Biosystems 3730xl DNA analyser (ThermoFisher Scientific). Raw data files were imported using GeneMarker (SoftGenetics LLC), normalised based on the size standard LIZ500, and scored using customised panels as previously described (Anderson et al., 1999;Ruybal-Pesántez et al., 2017). All major peaks that were within the expected marker base pair (bp) range and were spaced at intervals corresponding to trinucleotide (3 bp) repeats were considered to be true alleles.
Any peak less than 33% of the primary peak (i.e., local max) for a locus was considered a minor allele and not interpreted as a true allele. Background noise was defined as any peak <200 fluorescent units (Anderson et al., 1999). These data were cleaned using R package base v. 3.5.0 (R Core Team, 2018) and then processed using TANDEM v. 1.09 (Matschiner & Salzburger, 2009), which is optimal to assign an allele to each trinucleotide microsatellite locus for each isolate. We combined data from the pre-and post-IRS surveys prior to binning alleles with TANDEM to ensure each survey could be compared accurately to each other. and TA42 was <75% pre-IRS and/or post-IRS (  Figure S1).
Isolates with one peak at all microsatellite loci were defined as "true" single-clone infections (MOI = 1). Isolates with two or more peaks at ≥1 loci were considered to be multiple-clone infections (MOI > 1) and the multilocus haplotypes were constructed using the predominant peak at each locus. The combined data set with the single-clone and multiple-clone infections was defined as the "all infections" data set ( Figure S1). Isolates with single-clone infections and isolates with a maximum of two peaks at any locus (i.e., MOI = 2) were defined as the "dominant infections" data set (Ruybal-Pesántez et al., 2017). The "dominant infections" data set is robust and accounts for possible confounding by multiple-clone infections while still maximising the sample size available for analysis (Anderson et al., 1999).

| Genetic diversity
Binned data files were processed manually in Microsoft Excel v.
16.30 to generate input files for the population genetics software packages, as described below. Patterns of genetic diversity were an- H e was calculated using the formula: Where p is the allele frequencies at a given locus and n is the number of observed alleles in each locus (Nei, 1978).
"Dominant" constructed haplotypes were used for these analyses. Allelic richness (R s ) was calculated using the R package  (Piry et al., 1999). BOTTLENECK was used to test if our populations experienced a bottleneck event, henceforth defined as a recent severe reduction in effective population size (N e ). For selectively neutral loci, A and H e result from an equilibrium between mutation and genetic drift . In non-bottlenecked populations that are near this "mutation-drift" equilibrium, the H e will equal the heterozygosity expected at mutation-drift equilibrium (H eq ) (i.e., H e = H eq ). If a bottleneck event has occurred, A decreases immediately but H e is briefly retained, becoming larger than H eq (i.e., H e > H eq ) (Cotton et al., 2018;Piry et al., 1999). This transient excess can be used to detect recent bottleneck events (within 2-4N e generations), while population expansions typically exhibit a heterozygosity deficiency (where H eq > H e ) (Branch et al., 2011;Piry et al., 1999). H eq is calculated from the observed A and tions. This is a sufficient time period for the bottleneck calculation from Cornuet and Luikart (1996) to be able to predict whether there was a genetic bottleneck .
Using 1000 simulations, the infinite alleles model (IAM) and stepwise mutation model (SMM) were run as recommended by BOTTLENECK and as reported in the literature (Anderson et al., 2000;Branch et al., 2011;Cornuet & Luikart, 1996;Jennison et al., 2015;Piry et al., 1999). The IAM posits that every mutation event generates a new allele that is independent from its progenitor (Selkoe & Toonen, 2006), while the SMM contends that there is an equal probability that a mutation adds or subtracts one or more repeat units at a fixed rate. The SMM process mimics DNA replication errors that generate mutations and allows for a mutation to an existing allele (homoplasy) (Ellegren, 2004;Ohta & Kimura, 1973). This results in fewer distinct allele states than the IAM for the same mutation rate (Cornuet & Luikart, 1996).
For the SMM, N e was calculated as: And for the IAM, N e was calculated by: Where H e is the mean expected heterozygosity across all loci and µ is the microsatellite mutation rate for P. falciparum: Age groups a [95% confidence interval, 6.98 × 10 −5 -3.7 × 10 −4 ] (Anderson et al., 2000).
The SMM has been identified as a more stringent model for microsatellite data and was ultimately used for analysis (Cornuet & Luikart, 1996;Piry et al., 1999). A onetailed Wilcoxon's sign rank test was used to detect heterozygosity excess using allele frequency data .
Furthermore, BOTTLENECK was used to compare the distribution of allele frequencies observed in a population to the distribution expected in a nonbottlenecked population .

| Multilocus linkage disequilibrium
The extent of inbreeding within and between populations was estimated using the R package poppr v. 2.7.1 (Kamvar et al., 2014). The standardised index of association ( r d) (Agapow & Burt, 2001) was used to estimate the extent of multilocus linkage disequilibrium (LD, i.e., the nonrandom association of alleles) and is based on the index of association (I A ) (Smith et al., 1993). I A has been shown to increase steadily with the number of loci, therefore the standardised form, r d, was used to account for the number of loci sampled (i.e., 10 microsatellite loci) to reduce bias. To uncover whether any patterns of

| Genetic relatedness
To investigate genetic relatedness between isolates we calculated the pairwise allele sharing (P AS ) statistic using only those isolates in the "dominant infections" data set that had complete infection haplotypes. Complete multilocus haplotypes were used to ensure that the denominator would be consistent for all comparisons.
P AS scores were calculated by: Where N AB is the number of alleles shared between two infection haplotypes and N L is the microsatellite number of microsatellite loci (i.e., 10) (Ruybal-Pesántez et al., 2017).
The resulting P AS score is represented as a ratio, ranging from 1 (i.e., clones) to 0 (i.e., unrelated). Any infection haplotype pairs with a P AS ≤ 0.25 would be considered "unrelated", 0.25 < P AS < 0.5 would be considered "half-siblings", and a P AS ≥ 0.5 would be considered "related" (i.e., siblings or recent recombinants). P AS comparisons were calculated between all possible infection haplotype pairs both within (i.e., pre-and post-IRS) and between (i.e., pre-vs. post-IRS) the survey time points investigated.
To visualise whether the "related" infection haplotypes were clustered geographically, we constructed spatial genetic relatedness networks at a threshold of P AS ≥ 0.70 (i.e., identical at ≥7 of the 10 microsatellite loci). This threshold was selected to visualise the genetic similarity between isolates that probably resulted from a recent transmission and/or recombination event. These spatial networks were plotted using the R packages ggraph v.

TA B L E 1 (Continued)
the R package DEMEtics v. 0.8-7 (Gerlach et al., 2010). Jost's D is a heterozygosity-based estimator of population differentiation and the evolution of genetic divergence between populations (Gerlach et al., 2010;Jost et al., 2018). This measure is robust when the number of observed alleles per locus is greater than two and when the allelic diversity for a locus (within-population diversity) is high (Gerlach et al., 2010). , which can be used to determine the optimal K cluster (Pritchard et al., 2000). A second order rate of change of LnP[D], ΔK, was calculated according Evanno et al. (2005), which has been found to be a more sensitive method of predicting the real number of clusters. STRUCTURE Harvester v. 0.6.94 (Earl & vonHoldt, 2012) was used to process the results and calculate the optimal number of clusters from the peak ΔK according to the Evanno et al. (2005) method. CLUMPAK (Clustering Markov Packager Across K) v. 1.1 (Kopelman et al., 2015) was used to account for the number of stochastic simulations per K and to visualise the results at using time and spatial levels.

| Microsatellite study population
Using the 10 microsatellites markers, a subset of 192 and 200 participants with microscopic P. falciparum infections from the preand post-IRS surveys, respectively, were used for the microsatellite genotyping (Table 1). This subset of isolates selected based on MOI (see Section 2) was not statistically different than those isolates excluded for any of the key variables (p > .05) (Table S4), except for age pre-IRS (χ 2 = 15.46, p < .001) and parasitaemia post-IRS (p < .01). For the subset of participants selected, there were no significant differences between the pre-and post-IRS surveys for any of the demographic or parasitological parameters, except for age and reported antimalarial usage (p ≤ .002, Table 1). These agespecific differences were in the youngest (1-5 years) and oldest (6-10 years) children's age groups, such that there were fewer 1-5 year olds included post-IRS compared to pre-IRS (8.5% vs. 18.8% respectively) and a greater number of 6-10 year olds included post-IRS compared to pre-IRS (34.0% vs. 19.3%, respectively) ( Table 1).
These proportional differences, although significant, reflect the underlying epidemiological changes in the prevalence of P. falciparum infections by age following the IRS intervention with fewer 1-5 year olds being infected post-IRS (Tables S3 and S4). For antimalarial drug usage in the previous two weeks, similar to the patterns seen in the Bongo study population (Table S1), we found that reported antimalarial usage significantly declined following the IRS intervention (43.2% vs. 11.5%, pre-vs. post-IRS, respectively, p < .001) ( Table 1) Table 1.  Tables S5 and S9 for additional information)

| Microsatellite genetic diversity of P. falciparum
Based on microsatellite genotyping, 53.6% and 54.0% of the isolates in the pre-and post-IRS surveys, respectively, had an MOI > 1, indicating the majority of population in Bongo harboured multiple-clone infections pre-and post-IRS (Figure 2). In Bongo, high levels of genetic diversity were observed pre-and post-IRS for both the "all infections" and "dominant infections" data sets (Table S5 and Table 2) . Since measures of diversity were comparable between these data sets, the "dominant infections" data set was used for all the subsequent analyses to maximise the sample size available for analysis while minimizing the possible confounding effects of including multiclonal infections (i.e., "all infections" data set) (see Section 2; Tables S6-S8). The dominant infections" data set includes 128 and 156 isolates pre-and post-IRS, respectively (Table 2, Figure S1). In both the pre-and post-IRS surveys there was considerable allelic variation defined using both the number of alleles per locus (A) and allelic richness estimates (R s ) ( Table 2). There was a mean of 11.5 and 12.4 alleles per locus pre-and post-IRS, respectively, and a comparable estimated R s of 11.4 pre-IRS and 11.6 post-IRS (Table 2).
We also observed similar trends in A and R s for both catchment areas during the pre-and post-IRS surveys (Table 2). In addition, despite the observed reduction in transmission intensity following the IRS intervention (as measured by the entomological inoculation rate (EIR), see Supporting Information Methods) there were no significant changes in the population genetic measures of allelic variation described above (Table 2). We also observed similar trends in A and R s for both catchment areas during the pre-and post-IRS surveys, as well as no significant changes in these parameters pre-to post-IRS (Table 2).
The same number of haplotypes as isolates was observed pre-IRS, thus every multilocus haplotype was unique (Table 2). Post-IRS every haplotype was also unique, except for two individuals from the same compound (i.e., household) in Vea/Gowrie that shared the same infection haplotype (Table 2). In Bongo, H e remained high post-

IRS despite the IRS intervention with a significant, but slight increase
in H e between in the pre-and post-IRS surveys (pre-IRS = 0.79 vs. post-IRS = 0.81, p = .048, Table 2, Tables S5 and S9). This significant difference however was not maintained when we stratified by catchment area, indicating potential temporal differences.

| N e estimates and population bottleneck investigation
Effective population size (N e ) was calculated to estimate the seeding parasite populations in Bongo. N e increased pre-to post-IRS for both the stepwise mutation model (SMM) and infinite alleles model (IAM) ( Table 3, Tables S10 and S11 (Table 4; Tables S10 and S11). This is consistent with our findings above where the number of alleles (A), and therefore H e , was maintained at the levels observed before the introduction of the IRS intervention in Bongo. This suggests that there was no bottleneck event between the pre-and post-IRS surveys investigated (i.e., T1 and T2).

| Multilocus linkage disequilibrium
The index of association, r d, was used to assess multilocus linkage disequilibrium (LD), or nonrandom associations among loci using

| Genetic relatedness of P. falciparum infection multilocus haplotypes
Pairwise allele sharing (P AS ) comparisons between infection haplotypes were next used to examine parasite genetic relatedness. Even though all the multilocus haplotypes were unique (i.e., not repeated, except for one pair as discussed), it is possible that these infection haplotypes may only differ at one or two loci and may reflect  Table S12). When the infection haplotypes were compared over time between the pre-and post-IRS surveys, they were all highly unrelated (median P AS = 0.2), indicating that parasite clones (as defined by their multilocus microsatellite haplotypes) did not appear to be maintained temporally in Bongo ( Figure 5, Table S12). In fact, of the 6,804 pairwise comparisons between (i.e., pre-vs. post-IRS) the time point surveys, only 172 isolate pairs (2.5%) were "related" (i.e., siblings or recent recombinants) having haplotypes that shared ≥5 loci out of 10 loci (i.e., P AS ≥ 0.5), and just three isolate pairs (0.04%) were "highly related" being identical at ≥7 out of 10 loci (i.e., P AS ≥ 0.7) ( Figure 5, Table S12). Multilocus haplotype pairs were significantly less related post-IRS than pre-IRS (Wilcoxon test, p < .001, Table 6), suggesting that there was a reduction in sexual recombination/outcrossing following the IRS intervention.

| Pairwise allele sharing networks
To visualize these P AS relationships, we constructed spatial networks of the P AS scores between infection haplotypes to investigate if the "highly related" isolates (P AS ≥ 0.7) clustered at different spatial scales (i.e., at the level of catchment areas and/or households pre-or post-IRS) and/or spatiotemporally (i.e., pre-vs. post-IRS). We observed no apparent geospatial clustering and very few "highly related" haplotypes within pre-and post-IRS (i.e., T1 and T2, Figure 6a,b), or between pre-vs. post-IRS (i.e., T1 vs. T2, Figure 6c). Even when we Note: N e estimates are based on mean effective heterozygosity and estimated P. falciparum microsatellite mutation rate 1.59 × 10 −4 [95% confidence interval: 6.98 × 10 −5 -3.7 × 10 −4 ] (Anderson et al., 2000). The upper and lower confidence intervals for the mutation rate used to estimate N e are labelled in parentheses.
Abbreviations: IAM, infinite alleles model; SMM, stepwise mutation model. p-values for each population obtained from 9999 permutations of the data are labelled in brackets.

TA B L E 3 Effective population size (N
F I G U R E 3 Distribution of allele frequencies for the microsatellite loci genotyped pre-IRS (T1, October 2012) and post-IRS (T2, October 2015) using the "dominant infections" data set. The Jost's D and G ST values have been provided for each locus, along with the number of isolates that were genotyped per locus. For the number of isolates (N) with data for each locus, please see Table S6 for more details relaxed our threshold to examine "related" (P AS ≥ 0.5) haplotypes, there was no discernible spatial clustering.

| Genetic differentiation and fixation
To further investigate spatial patterns of genetic differentiation, we calculated the pairwise G ST and Jost's pairwise index of differentiation (D) over time (i.e., pre-to post-IRS) and over the different spatial scales (i.e., Bongo and the catchment areas) (Figure 7a). Both G ST and pairwise Jost's D were calculated to measure the extent of allelic fixation and allelic differentiation, respectively (Figure 7a), where Jost's D values can be interpreted as the mean proportion of "private alleles" between populations. Allele fixation differences were also found when we compared the allele distributions for each locus independently ( Figure 3).
When the pre-and post-IRS surveys were compared, the parasite population in Bongo was genetically differentiated (Jost's D: p = .034), with significant differences in allelic fixation (G ST : p = .021) (Table S13). In fact, 3.3% [95% CI: 2.1%-4.7%] of alleles across all 10 loci were considered private between the pre-and post-IRS surveys (Table S13). To determine if this temporal genetic differentiation was due to geospatial factors, we next stratified this analysis by catchment area. Using this approach there was clear evidence of spatial genetic differentiation between Vea/Gowrie and Soe, however this was only significant post-IRS (Jost's D: p < .01, Figure 7a,  Table S12) lower diagonal; Figure 7b). When we investigated this further, there was no evidence of genetic differentiation in Vea/Gowrie pre-to post-IRS (Jost's D: p ≥ .05, Figure 7a, lower diagonal; Figure 7c). In contrast, Soe, which shares an immediate border with the Nahouri Province in Burkina Faso to the north, was found to be genetically differentiated pre-to post-IRS (Jost's D: p < .01, Figure 7a, lower diagonal; Figure 7d). This result indicates that the parasite population in Soe post-IRS was genetically different from the population that was surveyed pre-IRS. There was no clear trend in G ST at the catchment level (Figure 7a, upper diagonal) therefore we could not attribute the significant G ST to a specific area and this may reflect the overall differentiation (Jost's D values).

| Bayesian cluster analysis
To further explore the observed spatiotemporal population structure, a Bayesian cluster analysis was performed. This infers the number of ancestral genetic clusters and assigns each infection haplotype to these clusters. The optimal number of genetic clusters in Bongo and after stratifying by catchment area were identified as K = 6 and 7, respectively (pre-to post-IRS, Figure S4). The higher K clusters observed from the pre-to post-IRS survey suggests that there were more distinct ancestral haplotypes over time (see Section 2). Nonetheless, all haplotypes segregated into evenly distributed genetic clusters ( Figure S5) and there was no evidence of P. falciparum subpopulation genetic clustering at any of the temporal or spatial scales investigated (i.e., populations were considered well-mixed).

| DISCUSS ION
Genomic surveillance of putatively neutral variation in microsatellites has proven informative to show that the short-term IRS intervention against a background of widespread LLIN usage in Bongo, did not bottleneck Plasmodium falciparum population diversity and structure.
Despite a more than 90% reduction in local transmission and a 35.7% reduction in parasite prevalence pre-to post-IRS, we observed that both genetic diversity and the effective population size of the parasite population increased slightly, rather than reduced. These data    Miller et al., 2002;Scherf et al., 2008). Our results suggest that the remaining diversity at the end of each dry season is sufficient to maintain the diversity of the parasite population post-IRS. Furthermore, our data point to gene flow from neighbouring uncontrolled areas as a potential source of increasing diversity. The observed increase in diversity may also be due to reduced outcrossing rates, leading to fewer "related" parasites (i.e., siblings or recent recombinants), and thus higher apparent genetic diversity in the parasite population. In fact, when we examined those isolates collected after the IRS intervention, we observed that they were significantly less likely to be "related" (i.e., limited to no sharing of alleles) compared to those isolates collected pre-IRS.
Of note, regardless of the intervention, all the multilocus haplotypes constructed for the dominant infections were unique (except for one pair of post-IRS isolates from the same compound (i.e., household) in Vea/Gowrie, Figure 6b) and virtually all isolate pairs (97.3%) shared less than 50% of their alleles. Moreover, a median of 20% of alleles were shared between haplotypes (i.e., highly unrelated), which was the same as previously observed at the end of the dry season in Bongo (Ruybal-Pesántez et al., 2017). This confirms our linkage equilibrium findings that IRS has not significantly perturbed outcrossing between unrelated or genetically diverse parasite clones. Moreover, the spatiotemporal P AS networks and Bayesian clustering analysis confirmed a lack of clustering among parasites, further supporting free gene flow and a lack of inbred clonal or highly related haplotypes persisting in Bongo.
Typical of high-transmission areas, there was no evidence of significant multilocus LD both within and between the pre-and ing those with perturbations such as LLINs and ACTs ( Carter et al., 2015;Chenet et al., 2012;Kattenberg et al., 2020;Roh et al., 2019).
Some pursuing multiple single nucleotide polymorphisms (SNPs) or microhaplotypes for surveillance may speculate that additional loci could increase resolution to detect changes in LD. This is certainly true in low-transmission. However, we believe our microsatellite result is robust, even with 10 loci in this high-transmission African setting due to high allelic diversity and extensive haplotype variation.
The high prevalence of multiclonal infections present in hightransmission areas of sub-Saharan Africa, like Bongo, limit the opportunity to sample the entire parasite population. Given we did not find a significant reduction, but a slight increase in diversity, it would be reasonable to assume that we have obtained an accurate picture even though restricting sampling to P. falciparum isolates with MOI ≤ 2, while excluding those with higher MOI. To date, there is no way around this restricted use of data in high-transmission settings.
Microsatellites do, however, provide a greater capacity to detect a higher number of alleles in the population with relatively few markers so can more accurately capture MOI up to three. Thus, for studies of genetic diversity and population structure in high-transmission settings, they can be preferable to using multiple SNPs that are only biallelic (Ellegren, 2004;Selkoe & Toonen, 2006). since Vea/Gowrie and Soe were genetically differentiated from Vea/ Gowrie post-IRS, with approximately 6.8% of alleles being private ( Figure 7a). However, given our findings of linkage equilibrium and lack of clustering, another explanation for these results may be due to the proximity of Soe to Burkina Faso, which borders Bongo District to the north (Figure 1a,c). Given that Burkina Faso has a high incidence of malaria, high parasite genetic diversity, and no significant population structure (De Allegri et al., 2013;PMI, 2017;Sondo et al., 2019), human occupational movement and/or vector migration may contribute to continuous gene flow and/or mixing of genetically diverse parasites between these two areas. Since no IRS was undertaken in the Nahouri Province proximal to Soe, the rate of importation of diverse genomes from Burkina Faso is expected to have remained the same over time. The significant reductions in P. falciparum prevalence in Soe following the IRS intervention means that these imported infections from Burkina Faso are more likely to be represented in our sampled population. Such observations have been reported within regions of sub-Saharan Africa (Bei et al., 2018;Duffy et al., 2017;Lynch & Roper, 2011;Mobegi et al., 2012;Roh et al., 2019;Sharp et al., 2007). Moreover, the implementation of control interventions in Burkina Faso has been limited to only 10%-40% population-wide distribution of LLINs between 2010-2014 (Samadoulougou et al., 2017). In the province that share an immediate border with Bongo District, no IRS programmes have been undertaken (PMI, 2017), therefore, vector and parasite populations in this region were not subjected to the same selection pressures as those in Bongo.

Consistent with our previous study undertaken in
These results indicate that, in high-transmission settings, short-term IRS interventions alone will not be sufficient to reduce parasite diversity. IRS will probably need to be sustained or combined with chemotherapeutic interventions (e.g., mass drug administration) to achieve low-transmission and bottleneck the parasite population in the human population in high-transmission areas like Bongo. Further investigation into these trends would provide insight into the threshold needed for perturbations to have a substantial effect on neutral diversity in these hightransmission settings where the P. falciparum reservoir remains highly diverse (He & Pascual, 2020). Our data show that future vector control interventions conducted in high-transmission settings in sub-Saharan Africa, will benefit by incorporating molecular surveillance to assess progress towards achieving the WHO Global Technical Strategy for Malaria 2016-2030 targets (World Health Organization, 2015a).

ACK N OWLED G EM ENTS
We wish to thank the participants, communities, and the Ghana Health Service in Bongo District, Ghana for their willingness to participate in this study. We would like to thank the field teams for their technical assistance in the field and the laboratory personnel at the Navrongo Health Research Centre for sample collection and parasitological/entomological assessments.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study have been made openly available in Dryad at https://doi.org/10.5061/dryad.kh189