Landscape attributes governing local transmission of an endemic zoonosis: Rabies virus in domestic dogs

Abstract

Landscape heterogeneity plays an important role in disease spread and persistence, but quantifying landscape influences and their scale dependence is challenging. Studies have focused on how environmental features or global transport networks influence pathogen invasion and spread, but their influence on local transmission dynamics that underpin the persistence of endemic diseases remains unexplored. Bayesian phylogeographic frameworks that incorporate spatial heterogeneities are promising tools for analysing linked epidemiological, environmental and genetic data. Here, we extend these methodological approaches to decipher the relative contribution and scale-dependent effects of landscape influences on the transmission of endemic rabies virus in Serengeti district, Tanzania (area ~4,900 km² ). Utilizing detailed epidemiological data and 152 complete viral genomes collected between 2004 and 2013, we show that the localized presence of dogs but not their density is the most important determinant of diffusion, implying that culling will be ineffective for rabies control. Rivers and roads acted as barriers and facilitators to viral spread, respectively, and vaccination impeded diffusion despite variable annual coverage. Notably, we found that landscape effects were scale-dependent: rivers were barriers and roads facilitators on larger scales, whereas the distribution of dogs was important for rabies dispersal across multiple scales. This nuanced understanding of the spatial processes that underpin rabies transmission can be exploited for targeted control at the scale where it will have the greatest impact. Moreover, this research demonstrates how current phylogeographic frameworks can be adapted to improve our understanding of endemic disease dynamics at different spatial scales.

Keywords: domestic dog; endemic zoonotic disease; landscape heterogeneity; phylogeography; rabies; spatial diffusion.

Figure 1

Resistance surfaces for landscape attributes hypothesized to influence rabies virus movement in the Serengeti district. Host density and distribution (a–c), host movement (d–g) and host susceptibility influenced by vaccination (h–j). Block arrows indicate whether the attribute was considered a facilitator (green) or barrier (red) to viral movement

Figure 2

The spatial location and phylogenetic structure of 152 sequenced rabies viruses sampled from 2004 to 2013 within the Serengeti district, Tanzania. (a) The Serengeti district (red polygon) within Tanzania; (b) locations of sequenced rabies cases within the Serengeti district (grey polygon) with underlying topography (map tiles by Stamen Design, under CC BY 3.0. Data by OpenStreetMap, under ODbL.) and administrative boundaries from http://www.nbs.go.tz; (c) timescaled maximum clade credibility tree from a Bayesian phylogenetic reconstruction of whole‐genome sequences, with node posterior support >0.9 indicated by blue circles. The inset shows node density through time for the posterior set of trees, with >90% nodes occurring in the last 10 years. Maps drawn using R packages OpenStreetMap (Fellows & Stotz, 2016) ggmap (Kahle & Wickham, 2013) and maptools (Lewin‐Koh et al., 2012)

Figure 3

Using resistance distances to incorporate landscape heterogeneity into phylogeographic frameworks. Illustration of resistance surfaces assuming rivers (dark red) acts as barriers to RABV spread. Two approaches are used to incorporate resistances in discrete phylogeographic reconstructions: (a) locations of sequenced rabies cases are morphed in space using multidimensional scaling (MDS) and clustered according to a k‐means partitioning scheme (k = 3 shown). MDS cluster information is used to assign traits in a discrete trait phylogeographic reconstruction measuring viral lineage migrations and phylogeny–trait association; (b) locations are clustered according to geographic distances using k‐means partitioning and resistance distances between cluster centroids are used to parameterize a GLM extension of discrete phylogeographic diffusion. Bayesian model averaging is used to identify significant predictors of viral spread between centroids

Figure 4

Summarized results from discrete‐MDS phylogeographic models using landscape‐informed spatial clusters for reconstructed RABV movement in Serengeti district. A number of spatial scales were examined by subjecting RABV cases (n = 152) to different levels of partitioning (k), ranging from 3 to 15 clusters. (a) A heatmap representing the reduction in estimated viral lineage migrations relative to a null model (where only isolation by distance (IBD) was used to inform spatial clustering) at each k (horizontal axis) when each landscape attribute (vertical axis) informed the configuration of clusters. White cells represent no reduction or an increase in migrations (i.e., the null model was better), whereas shaded cells represent fewer migrations between attribute‐informed clusters compared to the null model (i.e., the attribute‐informed model was better). (b) The number of inferred migrations at each spatial scale when clusters were assigned randomly, according to IBD, or by roads (which showed the largest reduction in migrations relative to IBD at k = 3–6). (c) A heatmap representing the improvement in phylogeny–trait association according to an association index, AI, for landscape‐informed clusters relative to IBD‐informed clusters, with smaller AI values indicating stronger associations. (d) The inferred AI at each spatial scale when clusters were assigned randomly, according to IBD, or using dog presence (which had the strongest phylogeny–trait association at smaller values of k)