Geographic-genetic analysis of Plasmodium falciparum parasite populations from surveys of primary school children in Western Kenya

Abstract

Background. Malaria control, and finally malaria elimination, requires the identification and targeting of residual foci or hotspots of transmission. However, the level of parasite mixing within and between geographical locations is likely to impact the effectiveness and durability of control interventions and thus should be taken into consideration when developing control programs. Methods. In order to determine the geographic-genetic patterns of Plasmodium falciparum parasite populations at a sub-national level in Kenya, we used the Sequenom platform to genotype 111 genome-wide distributed single nucleotide polymorphic (SNP) positions in 2486 isolates collected from children in 95 primary schools in western Kenya. We analysed these parasite genotypes for genetic structure using principal component analysis and assessed local and global clustering using statistical measures of spatial autocorrelation. We further examined the region for spatial barriers to parasite movement as well as directionality in the patterns of parasite movement. Results. We found no evidence of population structure and little evidence of spatial autocorrelation of parasite genotypes (correlation coefficients <0.03 among parasite pairs in distance classes of 1km, 2km and 5km; p value<0.01). An analysis of the geographical distribution of allele frequencies showed weak evidence of variation in distribution of alleles, with clusters representing a higher than expected number of samples with the major allele being identified for 5 SNPs. Furthermore, we found no evidence of the existence of spatial barriers to parasite movement within the region, but observed directional movement of parasites among schools in two separate sections of the region studied. Conclusions. Our findings illustrate a pattern of high parasite mixing within the study region. If this mixing is due to rapid gene flow, then "one-off" targeted interventions may not be currently effective at the sub-national scale in Western Kenya, due to the high parasite movement that is likely to lead to re-introduction of infection from surrounding regions. However repeated targeted interventions may reduce transmission in the surrounding regions.

Keywords: genotyping; heterogeneity; malaria; micro-epidemiological; parasite mixing; school surveys; spatio-temporal; western Kenya.

Figure 1.. Spatial distribution of primary schools surveyed in a geographically defined region of western Kenya.

2486 Plasmodium falciparum positive samples were collected from children in 95 primary schools in 20 districts in this region. ( A) Each dot represents an individual school, colour-coded by the administrative district in which it is located. ( B) Map showing a close-up of the study region.

Figure 2.. Distribution of Plasmodium falciparum positive samples and their associated genotyping success rates.

( A) 2486 samples were collected from 95 schools (33 in western province and 62 in Nyanza province) in western Kenya. The total number of samples varied from 1 to 81 per school. ( B) 111 single nucleotide polymorphic (SNP) positions were genotyped in all parasite positive samples. Mean genotyping success rates per school ranged from 6–86%.

Figure 3.. Spatial distribution of scores for the first 3 principal components (PCs) representing parasite genotypes.

Geospatial positioning information was collected at the school level, thus PC scores (values of the transformed variables corresponding to a specific data point) were aggregated for all samples in an individual school. Here each dot represents a school, and has been colour-coded based on the mean genotype score of all parasite isolates collected in that school. Cumulatively, the first three PCs accounted for only 10.78% of the variation observed in the genotype data (PC1=3.74%, PC2=3.54%, PC3=3.5%).

Figure 4.. Moran’s I correlation coefficients describing the spatial autocorrelation of genotypes of Plasmodium falciparum parasite pairs.

Spatial autocorrelation was tested separately for parasites grouped into three distance classes of a) 1km, b) 2km and c) 5km. Within each distance class, correlations were computed for each of the first 3 principal components (PCs). The asterisks represent those distances at which statistically significant (p<0.01) correlation coefficients were found for parasite pairs within each distance class, indicative of possible clustering of specific parasite genotypes.

Figure 5.. Spatial scan statistics to identify local spatially autocorrelated clusters of genetically distinct Plasmodium falciparum parasite sub-populations in western Kenya.

Spatial scan statistics employing the use of multiple circular windows of varying sizes (ranging from covering only 1 sample up to 50% of the sample population) around samples geographically defined regions was used to compute the ratio between expected and observed number of genotypes within each window. Each window with higher than expected number of similar genotypes was noted down as a cluster, and its statistical significance determined after accounting for the multiple comparisons. Genotypes for individual parasites were assigned based on scores of the first 3 principal components. Here, each school is colour-coded based on the mean principal component score for all parasite genotypes found within it. One cluster of highly related parasite genotypes (blue circle) was identified when analysing the second principle component.

Figure 6.. Variation in Plasmodium falciparum parasites’ genetic diversity over distance.

Genetic diversity was defined as the average number of single nucleotide polymorphism (SNP) differences between parasites in each pairwise school comparison, and was plotted against the distance between the corresponding school pair. The blue line represents loss-fitted smoothing with 95% confidence intervals (grey area).

Figure 7.. Raster analysis by pixels to examine the presence of spatial barriers to Plasmodium falciparum movement in a geographically defined region of western Kenya.

( A) Each pixel represents a 10km-by-10km area of the region, and is colour-coded based on the coefficient estimates derived from a linear regression analysis that was used to test the impact of each pixel in acting as either a barrier (blue pixels) or gateway (red pixels) to parasite movement with the region. No pixels were significant barriers or gateways to parasite movement after Bonferroni correction to account for multiple comparisons. ( B) Distribution of p values observed after bootstrapping the regression analysis (with 10,000 resampling steps) to determine the level of significance of pixels in acting as barriers to parasite movement. ( C) Moran’s I analysis describing the spatial autocorrelation between geographical locations of pixels and their associated coefficient estimates. Autocorrelation was calculated for parasites grouped in 10km distance bands, and the analysis was bootstrapped 100 times to determine significance.

Figure 8.. Raster analysis by pixels to examine patterns of north/south versus east/west directional movement of Plasmodium falciparum parasites in western Kenya.

( A) Each pixel represents a 10km-by-10km area of the region, and is colour-coded based on coefficient estimates describing the effect size of each pixel in influencing directional movement. Pixels that were statistically significant after correcting for multiple testing are highlighted with black borders. Grids were colour-coded to represent east/west (red) or north/south (blue) movement. ( B) Distribution of p values observed after bootstrapping the regression analysis (with 10,000 resampling steps) to determine the level of significance of pixels in influencing parasite directional movement. ( C) Moran’s I analysis to describe the spatial autocorrelation of movement within the region. The analysis was computed using geographical coordinates of individual pixels to represent feature locations and coefficient estimates derived from the bearing regression analysis to represent the associated feature values. Autocorrelation was computed for parasites grouped in 10km distance bands. Significant positive correlation coefficients (p<0.01; marked by asterisks) were observed for schools separated by up to 40 km within the 10km distance bands.

Figure 9.. Map of the western Kenya study area with raster grids representing bearing analyses superimposed on top of it.

Multivariable linear regression analysis was carried out to determine bearing (directionality of movement) of Plasmodium falciparum parasites among schools in the region. Grids are colour coded based on the coefficient estimates describing the effect size of that grid in influencing directional movement. Red represents east/west movement, while blue represents north/south movement. The grids with black borders represent those areas that were significant in east/west movement, even after Bonferroni-correction for multiple testing. The blue circle shows the region of the study site that had predominantly north/south movement, while the red circle represents that region that had predominantly east/west movement. Each dot represents a school, colour-coded based on the district in which the school is located.