Our model describes how vector–host contact heterogeneity owing to preferential mosquito feeding drives WNV enzootic transmission dynamics at four different locations in CT. Site-specific mosquito and bird abundances were represented in the model. From field data we identified host preferences for American robins by Cx. pipiens
using the feeding index (αv
) and with our model showed that this is the most influential parameter on both WNV transmission intensity and timing. This finding is consistent with host preferences by Cx. pipiens
for American robins demonstrated by both field studies [7
] and experimental trials [29
]. As a consequence, it has been suggested that preference-induced contact heterogeneity [7
] may influence pathogen transmission dynamics. Our model shows that strong host-feeding preference by Cx. pipiens
for American robins is possibly a major driver for WNV transmission in CT.
Our model advances previous modelling efforts because it is both parametrized and validated using field data. We also used a parsimonious modelling approach: rather than modelling each site independently, we developed a single transmission model and subsequently incorporated site-specific field data from various locations of interest. This provides the advantage of a flexible model for exploring transmission dynamics both within and between multiple localities.
We identified a feeding preference threshold necessary for maintaining Cx. pipiens-mediated transmission of WNV (αv > 6). Increasing the degree of feeding on American robins resulted in more intense transmission occurring earlier in the season. The greatest influence was observed when preference was moderate (6 < αv <20); when αv > 20, transmission intensity and timing approached a plateau. This finding is consistent with how the threshold quantity Ro depends on αv (see the electronic supplementary material). Our model predicts maximum enzootic transmission potential at each site based on mosquito abundance and the relative abundance of preferred versus non-preferred hosts (). The level of host preference by Cx. pipiens, measured by αv, subsequently determines the timing and intensity of transmission and whether the plateau is reached. Inter-site differences in host preferences may be driven by landscape structure influencing the microhabitat availability of hosts to mosquitoes, availability of alternative hosts, etc; further research at a finer spatial scale may be warranted.
We used the VI to qualitatively validate that the model provides realistic estimates of infection in Cx. pipiens. The model agreed well with site-specific VI at sites A–C, predicting high densities of infected Cx. pipiens at site C, moderate densities at site B, and low densities at site A, but no agreement was observed between model and field data for site D. Lack of fit between model predictions and VI could be owing to simplified model assumptions, such as stable bird population densities, only two avian species represented, or similar infectivity and transmissibility for both avian hosts. On the other hand, the VI estimate is limited by field sampling biases and parameter estimation errors and low sample sizes, requiring using a monthly average of the VI. Unreliable IR estimates owing to small sample sizes for site D (38 pools tested, only 1 positive pool) may have led to the lack of fit with the model prediction.
Model results, along with IRs in Cx. pipiens
in our study, further support the role of Cx. pipiens
as the primary enzootic vector in this region. Culex pipiens
has been established as an important enzootic vector by consistent isolations of WNV from mosquito trap collections [11
], by its ornithophilic feeding behaviour [13
], and associations between virus-infected mosquitoes and dead bird reports [42
]. This species has also been incriminated as a bridge vector in Illinois [47
]. However, in our study, 94.65 per cent (139 out of 147) of blood meals examined were avian-derived and no human-derived samples were obtained, indicating that other mosquito species also serve as bridge vectors during epidemics in CT. Models to determine how feeding behaviours of bridge vectors influence epidemic transmission patterns would be useful to determine the risk of spill-over of WNV from birds to humans.
Our model builds on previously published epidemiological models that use generalized parameters to understand WNV transmission dynamics [27
]. Previous models were rarely field-validated, since many were not empirically informed. Field-derived parameters are difficult to measure precisely and labour-intensive to obtain, but are vital to understanding real-world transmission dynamics scenarios. In contrast to previous models [27
], we used raw counts from mosquito traps to derive the rate of change in abundance. Other models have used differing assumptions to parametrize the vector population (reviewed in [54
]). They may postulate some level of constant population growth [55
], use a step-function [27
], simulate growth over a season [51
], or estimate production in a compartmental model consisting of egg, larvae and emerging susceptible adult stages [49
]. By contrast, our method wholly captures the variation in the adult mosquito abundance data and does not require estimates for immature or other mosquito age classes.
WNV provides a suitable model for studying transmission dynamics of multi-host vector-borne pathogens because it is maintained in an enzootic cycle between mosquito vectors and multiple avian host species, with epizootic events in the United States every year. Given the strength of our findings, these results may be extended to other vector-borne diseases. A multitude of host-choice experiments provide convincing evidence that many malaria vectors, including Anopheles gambiae
, prefer humans over cattle [56
], for example. Culex nigripalpus
, a major vector for Saint Louis encephalitis in the southern United States, preferred chickens to bobwhite quail in a host-choice experiment in Florida, USA [64
]. Host preferences are not exclusive to mosquitoes: Triatoma
bugs that transmit Chagas' disease demonstrated strong feeding preferences for dogs over both chickens and cats in a recent study [65
]. At least one species of tsetse fly (Glossina morsitans morsitans
), vector of human (sleeping sickness) and animal trypanosomiasis in Africa, show feeding preferences towards buffalo and oxen but avoid other hosts such as waterbuck (reviewed in [66
]). Extension of this model to other systems could determine how host preferences influence transmission dynamics.
In this study, we identified vector host preferences as the most important transmission parameter and quantified the contribution of preference-induced contact heterogeneity to enzootic transmission. The ‘dilution effect’ [5
], whereby the presence of less pathogen-competent hosts in a more diverse community results in ‘wasted’ transmission events and therefore lower overall pathogen prevalence, relies on the assumption that vectors are generalists. Extensive application of the dilution effect to Lyme disease, for example, has been based on the assumption of no or low host preferences by ticks ([67
], but see [68
] for evidence of host preferences by ticks). Our study indicates that the host community composition experienced by the pathogen may be quite different from the host community we measured; incorporating host preferences and community reservoir competence into transmission models may be essential for modelling transmission dynamics and predicting epizootics.