Using spatial genetics to quantify mosquito dispersal for control programs

Abstract

Background: Hundreds of millions of people get a mosquito-borne disease every year and nearly one million die. Transmission of these infections is primarily tackled through the control of mosquito vectors. The accurate quantification of mosquito dispersal is critical for the design and optimization of vector control programs, yet the measurement of dispersal using traditional mark-release-recapture (MRR) methods is logistically challenging and often unrepresentative of an insect's true behavior. Using Aedes aegypti (a major arboviral vector) as a model and two study sites in Singapore, we show how mosquito dispersal can be characterized by the spatial analyses of genetic relatedness among individuals sampled over a short time span without interruption of their natural behaviors.

Results: Using simple oviposition traps, we captured adult female Ae. aegypti across high-rise apartment blocks and genotyped them using genome-wide SNP markers. We developed a methodology that produces a dispersal kernel for distance which results from one generation of successful breeding (effective dispersal), using the distance separating full siblings and 2nd- and 3rd-degree relatives (close kin). The estimated dispersal distance kernel was exponential (Laplacian), with a mean dispersal distance (and dispersal kernel spread σ) of 45.2 m (95% CI 39.7-51.3 m), and 10% probability of a dispersal > 100 m (95% CI 92-117 m). Our genetically derived estimates matched the parametrized dispersal kernels from previous MRR experiments. If few close kin are captured, a conventional genetic isolation-by-distance analysis can be used, as it can produce σ estimates congruent with the close-kin method if effective population density is accurately estimated. Genetic patch size, estimated by spatial autocorrelation analysis, reflects the spatial extent of the dispersal kernel "tail" that influences, for example, the critical radii of release zones and the speed of Wolbachia spread in mosquito replacement programs.

Conclusions: We demonstrate that spatial genetics can provide a robust characterization of mosquito dispersal. With the decreasing cost of next-generation sequencing, the production of spatial genetic data is increasingly accessible. Given the challenges of conventional MRR methods, and the importance of quantified dispersal in operational vector control decisions, we recommend genetic-based dispersal characterization as the more desirable means of parameterization.

Keywords: Close kin; Dispersal kernel; Genome-wide SNPs; IBD; Mosquito dispersal; Spatial autocorrelation.

Fig. 1

Sampling locations and density distributions of observed separation distances. Red dots indicate the vertical trapping locations in apartment buildings in Tampines (a) and Yishun (b). Horizontal violin plots (c) show the density distribution of separation distance for full siblings, 2nd-degree relatives, 3rd-degree relatives, all close kin combined (“all CK”), non-close kin (“all non CK”), null distribution (“non CK random”), and traps. The box within each violin plot shows the interquartile range and the location of the median

Fig. 2

Effective dispersal distance kernel estimated from the close-kin data. The inferred pdfs are highly congruent among separate datasets (full sibling, 2nd- and 3rd-degree relatives) and the combined dataset (“all CK”), and are significantly different from the randomly subsampled non-close kin dataset (“non CK random”) that represents the null distribution of distances for randomly spaced individuals across the matrix of traps

Fig. 3

Isolation-by-distance analysis on non-close kin data from Tampines and Yishun. Mantel test and linear regression were applied to the matrices of PCA genetic distance and linear geographic distance between pairs of individuals in Tampines (upper) and Yishun (lower), with close kin removed from both datasets. The red line shows regression with 95% CI (dashed lines)

Fig. 4

The dispersal kernel spread (σ) estimated from the close-kin data and IBD analysis. Sigma (σ) and its 95% CI are plotted for the combined close-kin data (CK method) and PCA-based IBD analysis for Tampines and Yishun (with effective density estimates from methods 1–3)

Fig. 5

Spatial genetic autocorrelation in Tampines and Yishun. The ending point of a distance class is on the x-axis, and spatial autocorrelation coefficient (r) of genotypes in Tampines (107 individuals) and Yishun (108 individuals) is on the y-axis. Two dashed lines along the x-axis are the permutated 95% CI of autocorrelations under the null hypothesis of a random distribution of genotypes in space. Vertical lines are the bootstrapped 95% CIs with the mean genetic autocorrelation