Estimating probability of infection
For every sampled individual (n = 323), we selected ASF-positive
individuals that could have been the source of infection using a
temporal window (30 days) that reflects the upper limit of lifespan
after ASF infection, i.e. infection-to-death time (Gallardo et al.,
2017; Pietschmann et al., 2015). For samples originating from carcasses,
the window was set to 60 days prior to sampling and included
infection-to-death time and carcass decomposition time during which the
sample was taken. A decomposition time of 30 days was chosen arbitrarily
because the decomposition status of most carcasses was unknown and the
exact date of death could not be determined. This time frame was
conservative given usually longer decomposition times of wild boar
tissues (Probst et al., 2020) and ASF virus persistence in them (Fischer
et al., 2020). Additionally, the temporal window was extended by 7 days
post-sampling of the focal individual to account for the retrieval and
reporting lag of the ASF-positives. ASF-positive individuals selected
within those time frames were then categorized into four distance
classes (0-2 km, 2-5 km, 5-10 km, >10 km). The proportion
of ASF cases in each distance class were used as covariates explaining
the binary response of ASF infection status of the focal individual (0 -
negative, 1- positive).
We analysed the effects of explanatory variables on the probability of
ASF infection using a generalized linear model with a binomial response
and ”logit” link function (Zuur, 2009). The probability of ASF infection
reflects the likelihood of any sampled individual from the study area
testing positive for ASF within the 2.5 years of the study period based
on the covariates. Proportion of ASF-positive individuals within the
four aforementioned distance classes and relatedness to ASF-positive
individuals within 0-10 km was used as covariates to verify the
predictions (Table 1). Sample sizes across covariate categories were
unbalanced and precluded formulation of a global model including all
explanatory variables. Therefore, the structure of the models was
developed progressively, starting with simple formulations where each
model included just one covariate explaining infection risk
(Supplementary Information: Table S1). We then contrasted, in four
separate models (Table 2), information on relatedness to ASF-positive
individuals within 0-10 km with the proportion of ASF-positive
individuals in each of the four distance classes. This approach allowed
us to test the relative effect of relatedness and spatial proximity to
infectees on infection risk while accounting for the greater number of
highly related individuals at close distances resulting from wild boar
social structure. Explanatory variables did not show significant
collinearity (variance inflation factors - VIFs < 1.3 for all
models). We did not perform model selection for reduced models within
each hypothesis as we were interested in testing specific predictions
represented by each of the candidate explanatory variables in our full
models and we could not compare across models because the response
variable was different for each model. We quantified variation explained
by the fixed effects of the model by calculating marginalR 2GLMM (Nakagawa et al., 2017)
using the MuMIn R-package (Bartoń 2020). All statistical analyses were
performed in R 4.0.2 (R Core Team 2020). The models were computed using
the ”lme4” package (Bates et al. 2015).