Co-occurrence probabilities between mosquito vectors of West Nile and Eastern equine encephalitis viruses using Markov Random Fields (MRFcov)

Authors: SALLAM, MOHAMED - University Of Nevada; WHITEHEAD, SHELLY - Manatee County Florida; BARVE, NARAYANI - University Of Florida; BAUER, AMELY - University Of Florida; GURALNICK, ROBERT - University Of Florida; ALLEN, JULIE - University Of Nevada; TAVARES, JASMIN - University Of Florida; Gibson, Seth; Linthicum, Kenneth - Ken; GIORDANO, BRYAN - University Of Florida; CAMPBELL, LINDSAY - University Of Florida

Submitted to: Parasites & Vectors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/22/2022
Publication Date: 1/10/2023
Citation: Sallam, M., Whitehead, S., Barve, N., Bauer, A., Guralnick, R., Allen, J., Tavares, J., Gibson, S., Linthicum, K., Giordano, B.V., Campbell, L. 2023. Co-occurrence probabilities between mosquito vectors of West Nile and Eastern equine encephalitis viruses using Markov Random Fields (MRFcov). Parasites & Vectors. 16:10. https://doi.org/10.1186/s13071-022-05530-1.
DOI: https://doi.org/10.1186/s13071-022-05530-1

Interpretive Summary: The risk of transmission of mosquito borne viruses such as eastern equine encephalitis virus (EEEV) and West Nile virus (WNV) to humans and livestock in the US is driven by the presence of competent mosquito vector populations in a given area. For both EEEV and WNV in the US state of Florida this risk is a composite distributed across multiple species of competent mosquito vectors that may be sympatric but vary in relative abundance and vector competence. To protect public and veterinary health from these dangerous zoonotic viruses mosquito control districts (MCDs) conduct routine surveillance for these vector species to guide specific vector control efforts. Unfortunately, vector control capabilities are limited and need to be carefully targeted in space and time to be effective. Importantly, vector control techniques vary depending on the target species which adds to the complexity of designing effective control programs against mosquito vectors of EEEV and WNV. In this study we investigate for the first time the capability of a conditional Markov Random Fields technique to determine whether the relative abundances of species in a community of EEEV- and WNV- mosquito vector species in Manatee County, Florida, can be predicted by abiotic (land cover) and/or biotic (the particular suite of species present and bioclimatic variables) factors. Products from this model could improve targeting of MCD vector control activities to better protect public and veterinary health in Florida.

Technical Abstract: Mosquito vectors of eastern equine encephalitis virus (EEEV) and West Nile virus (WNV) in the US reside within broad multi-species assemblages that vary in spatial and temporal composition, relative abundances, and vector competence. These variations impact the risk of pathogen transmission and the operational management of these species by local public health vector control districts. However, most models of mosquito vector dynamics focus on single species and do not account for co-occurrence probabilities between mosquito species pairs across environmental gradients. In this investigation, we use for the first time conditional Markov Random Fields (MRFcov) to evaluate spatial co-occurrence patterns between host-seeking mosquito vectors of EEEV and WNV around sampling sites in Manatee County, FL. Specifically, we aimed to 1) quantify dependencies between mosquito vector species and other mosquito vector and non-vector species, 2) quantify dependencies between mosquito vectors and landscape and climate variables, and 3) investigate whether the strength of dependencies between species pairs are conditional on landscape or climate variables. We hypothesized that either mosquito vector species co-occur with other species in patterns driven by the landscape and/or climate variables, or that landscape variables unconditionally predict species abundances individually. Results indicated that landscape and bioclimatic covariates did not substantially improve the overall model performance and that the log abundances of the majority of WNV and EEEV vector species were positively dependent on other vector and non-vector mosquito species, unconditionally, but only weakly dependent or not at all dependent on environmental variables with one exception, Culiseta melanura, the primary vector for EEEV. Culiseta melanura showed a strong dependency on cropland and precipitation seasonality but not other species. Our analyses showed that some of the mosquito vector species may be habitat generalists, indicated by unconditional dependency, which could have confounded our analysis, but also indicated that the approach could be operationalized to leverage species co-occurrences as indicators of vector abundances in unsampled areas, or under scenarios where environmental variables are not informative.