Effects of host extinction and vector preferences on vector-borne disease risk in phylogenetically structured host-hector communities

Abstract

Anthropogenic disturbance impacts the phylogenetic composition and diversity of ecological communities. While changes in diversity are known to dramatically change species interactions and alter disease dynamics, the effects of phylogenetic changes in host and vector communities on disease have been relatively poorly studied. Using a theoretical model, we investigated how phylogeny and extinction influence network structural characteristics relevant to disease transmission in disturbed environments. We modelled a multi-host, multi-vector community as a bipartite ecological network, where nodes represent host and vector species and edges represent connections among them through vector feeding, and we simulated vector preferences and threat status on host and parasite phylogenies. We then simulated loss of hosts, including phylogenetically clustered losses, to investigate how extinction influences network structure. We compared effects of phylogeny and extinction to those of host specificity, which we predicted to strongly increase network modularity and reduce disease prevalence. The simulations revealed that extinction often increased modularity, with higher modularity as species loss increased, although not as much as increasing host specificity did. These results suggest that extinction itself, all else being equal, may reduce disease prevalence in disturbed communities. However, in real communities, systematic patterns in species loss (e.g. favoring high competence species) or changes in abundance may counteract these effects. Unexpectedly, we found that effects of phylogenetic signal in host and vector traits were relatively weak, and only important when phylogenetic signal of host and vector traits were similar, or when these traits both varied.

Fig 1. Overview of the modelling approach.

The schematic shows the four main steps in the model, labelled 1–4. Note that λ is adjusted to reflect variable phylogenetic signal in the simulated traits, which assumes a Brownian motion model of evolution.

Fig 2. Bipartite networks for representative simulations with small and large values of host specificity (g).

Width of bar reflects overall summed preference by vectors (top, taxa labeled starting with v) for hosts (bottom, taxa labelled starting with h), and plots show all preference scores between vectors and hosts that are 0.01 or greater (maximum value is always 1). The smallest version of g used was 0.021 (top plot), revealing generalist vectors. The largest value of g was 4.98, with more vectors specializing on only a few hosts.

Fig 3. Effect of host specificity on Q.

Simulations in which g was varied, holding all other parameters constant at their values in Table 1. A strong positive association was found (see text and Table 4).

Fig 4. Association between phylogenetic signal (λ) and modularity (Q₀).

Results are from bivariate tests, with λ_h1 = λ_v.

Fig 5. Modularity before and after extinction.

Line shows a slope of 1 through the intercept; thus, positive residuals indicate higher Q after extinction (i.e., Q₁ > Q₀, and ΔQ is negative). Results here are from simulations that varied phylogenetic signal in the vector trait (v), but similar results were found in other simulations.

Fig 6. Relationship between host specificity (g) and modularity (Q).

Data come from the 1000 simulations using the Latin hypercube sample (LHS). The measure of modularity is pre-extinction (i.e., Q₀).

Fig 7. Effects of variables on maximum and minimum values of Q.

For each variable, we indicate the 50 highest (+ symbol) and 50 lowest (circles) values of Q₀ from simulations conducted with the LHS. This provides a way to assess whether high extreme output values are consistently associated with particularly high or low values of each variable.

Fig 8. Extreme values of pre-extinction modularity—Post-extinction modularity (ΔQ) in relation to eight parameters in the LHS.

For each variable, we indicate the 50 highest (+ symbol) and 50 lowest (circles) values of ΔQ from simulations conducted with the LHS. This provides a way to assess whether high extreme output values are consistently associated with particularly high or low values of each variable. ΔQ is calculated as Q0 –Q1, such that negative values indicate positive changes in modularity following extinction.

Fig 9. Prevalence of infection in the simulated communities in relation to host modularity.

We ran simulations on 100 networks from a Latin Hypercube sample, using the S-I model described in the Methods. Results here represent prevalence at time step 300, averaged across simulations in which the disease was initiated in different hosts.