Quantifying connectivity between local Plasmodium falciparum malaria parasite populations using identity by descent

Abstract

With the rapidly increasing abundance and accessibility of genomic data, there is a growing interest in using population genetic approaches to characterize fine-scale dispersal of organisms, providing insight into biological processes across a broad range of fields including ecology, evolution and epidemiology. For sexually recombining haploid organisms such as the human malaria parasite P. falciparum, however, there have been no systematic assessments of the type of data and methods required to resolve fine scale connectivity. This analytical gap hinders the use of genomics for understanding local transmission patterns, a crucial goal for policy makers charged with eliminating this important human pathogen. Here we use data collected from four clinics with a catchment area spanning approximately 120 km of the Thai-Myanmar border to compare the ability of divergence (FST) and relatedness based on identity by descent (IBD) to resolve spatial connectivity between malaria parasites collected from proximal clinics. We found no relationship between inter-clinic distance and FST, likely due to sampling of highly related parasites within clinics, but a significant decline in IBD-based relatedness with increasing inter-clinic distance. This association was contingent upon the data set type and size. We estimated that approximately 147 single-infection whole genome sequenced parasite samples or 222 single-infection parasite samples genotyped at 93 single nucleotide polymorphisms (SNPs) were sufficient to recover a robust spatial trend estimate at this scale. In summary, surveillance efforts cannot rely on classical measures of genetic divergence to measure P. falciparum transmission on a local scale. Given adequate sampling, IBD-based relatedness provides a useful alternative, and robust trends can be obtained from parasite samples genotyped at approximately 100 SNPs.

Fig 1. SMRU clinics on the Thai-Myanmar border.

The border between Myanmar and Thailand (also the Moei river) is depicted in blue. Grey edges are proportional to inter-clinic proportions of highly related barcode parasite sample pairs (pairs with ${\widehat{\mathit{\pi }}}_{\mathbf{I}\mathbf{B}\mathbf{D}}>\mathbf{0.5}$). Latitudes and longitudes, respectively, are: 17.128107, 98.382152 (Maela); 16.83014, 98.53737 (Wang Pha); 16.5781479, 98.5846176 (Mae Kon Ken); 16.3258896, 98.670166 (Mawker Thai).

Fig 2. F_ST estimates based on 2001–2014 barcode data plotted with respect to inter-clinic distance.

Annotations refer to site comparisons using abbreviated clinic names (MLA for Maela, 212 parasite samples; WPA for Wang Pha, 457 parasite samples; MKK for Mae Kon Ken, 116 parasite samples; and MKT for Mawker Thai, 388 parasite samples). All parasite samples were single-infection. For a clinic pair, A and B say, the F_ST estimate was based on n_A + n_B parasite samples, where n denotes the number of parasite samples per clinic. Error bars represent 95% confidence intervals based on bootstrapping over SNPs.

Fig 3. F_ST estimates based on 2001–2014 WGS data plotted with respect to inter-clinic distance.

Annotations refer to site comparisons using abbreviated clinic names (MLA for Maela, 55 parasite samples; WPA for Wang Pha, 103 parasite samples; MKK for Mae Kon Ken, 4 parasite samples; and MKT for Mawker Thai, 16 parasite samples). All parasite samples were single-infection. For clinic pair, A and B say, the F_ST estimate was based on n_A + n_B parasite samples, where n denotes the number of parasite samples per clinic. Error bars represent 95% confidence intervals, based on boostrapping over SNPs.

Fig 4. Logit-transformed proportions of highly related 2001–2014 barcode parasite sample pairs with respect to inter-clinic distance.

Annotations refer to site comparisons using abbreviated clinic names (MLA for Maela, 212 parasite samples; WPA for Wang Pha, 457 parasite samples; MKK for Mae Kon Ken, 116 parasite samples; and MKT for Mawker Thai, 388 parasite samples). All parasite samples were single-infection. For inter-clinic pair A and B say, the proportion was based on n_A × n_B parasite sample pairs, where n denotes the number of parasite samples per clinic. Error bars represent 95% confidence intervals based on bootstrapping over highly related parasite sample pair labels (equal to 1 if ${\widehat{\mathit{\pi }}}_{\mathbf{I}\mathbf{B}\mathbf{D}}>\mathbf{0}.\mathbf{5}$ and 0 otherwise), and are therefore zero where there are no ${\widehat{\mathit{\pi }}}_{\mathbf{I}\mathbf{B}\mathbf{D}}>\mathbf{0}.\mathbf{5}$.

Fig 5. Logit-transformed proportions of highly related 2001–2014 WGS parasite sample pairs plotted with respect to inter-clinic distance.

Annotations refer to site comparisons using abbreviated clinic names (MLA for Maela, 55 parasite samples; WPA for Wang Pha, 103 parasite samples; MKK for Mae Kon Ken, 4 parasite samples; and MKT for Mawker Thai, 16 parasite samples). All parasite samples were single-infection. For inter-clinic pair A and B say, the proportion was based on n_A × n_B parasite sample pairs, where n denotes the number of parasite samples per clinic. Error bars represent 95% confidence intervals based on bootstrapping over highly related parasite sample pair labels (equal to 1 if ${\widehat{\mathit{\pi }}}_{\mathbf{I}\mathbf{B}\mathbf{D}}>\mathbf{0}.\mathbf{5}$ and 0 otherwise), and are therefore zero where there are no ${\widehat{\mathit{\pi }}}_{\mathbf{I}\mathbf{B}\mathbf{D}}>\mathbf{0}.\mathbf{5}$.

Fig 6. Proportions of significant negative spatial trend estimates with respect to data subset sample size.

Spatial trend estimates were based on regression of highly related parasite sample pair labels (equal to 1 if ${\widehat{\pi }}_{\mathrm{I}\mathrm{B}\mathrm{D}}>0.5$ and 0 otherwise) onto distance, within temporally adjusted models (β^adjusted ΔDistance (km) for barcode data, and β^{adjusted 2014} ΔDistance (km) for WGS data).