Biological, Environmental, and Anthropogenic Drivers
Due to insufficient sample size, data for A. geoffroyi andS. oerstedii were not included for further statistical analyses. Seven independent variables were explored for their association with exposure to T. gondii in A. palliata and C. imitator . Distance to villages, human population size, and human population density in the nearest village were considered indirect indicators of proximal domestic cats. The percentage of forest cover, annual average temperature, and annual precipitation were used as variables that affect oocyst survival. NP sex was used as an intrinsic factor for oocyst exposure (Table 1).
Forest cover, annual average temperature, and annual precipitation were measured within a 1.54 km2 circular area (700 m radius) that was constructed around each sample“s geolocation point (QGIS software, version 2.14). Data for these variables were obtained from Costa Rica Digital Atlas (2014) and WorldClim (Hijmans et al., 2005). The buffer size aimed to include both species“ home ranges and adjacent areas (Wainwright 2007).
Generalized linear models (GLM) with binomial distribution were run in R V3.4.2 (R Core Team, 2013) to explore associations between predictor variables and T. gondii seropositivity. The original predictors (forest cover and precipitation) were centered by subtracting their means and scaled by dividing by their standard deviations. The most parsimonious model was chosen by selecting the one with the lowest Akaike information criterion (AIC). Akaike weights (wi) were calculated to determine the evidence in favor of each model and estimate the relative importance of variables (Table 2).
Each NP species was modeled separately. Multicollinearity among the seven variables was evaluated by means of the variance inflation factor (VIF). When analyzing all the proposed variables, a very high VIF value (>10) was obtained suggesting multicollinearity, therefore the model was reduced to four variables (sex, forest cover, annual precipitation, and human population density). With these models, none of the variables obtained a VIF value greater than 10 (Table 2).