Analyses
From the amount of food each individual consumed in the food preference tests we derived metrics aimed at estimating niche centre, breadth and overlap (De Cáceres et al. 2011). We computed a distance matrix between food types based on their nutritional content (hydrocarbons, fat and proteins) while also taking into account the physical characteristics of each food type (length and width). These variables, standardized in order to remove differences due to units of measurement, were used to calculate the Euclidean distance between pairs of seed types. The niche centre in this resource space was computed by means of a principal coordinate analysis (PCoA), averaging the coordinates of the resources preferred by the individual in each trial (De Cáceres et al. 2011). The niche centre was defined as the multivariate mean in the subspace of the two first axes, which together accounted for > 80% of the observed variation. To describe niche breadth while accounting for resemblance among resources, we used Rao’s quadratic entropy as implemented in the ‘indicspecies’ package (De Cáceres & Legendre 2009). In addition, we also estimated the proportion of diet variation of each population that was due to either variation within (WIC) or among individuals (TNW). To this purpose, we modified the R package “indicspecies” to be able to deal with the fact that some resources are more similar among them than others (De Cácereset al. 2011). Finally, we estimated the mean pairwise niche overlap between each pair of individuals as a measure of the overall similarity among individuals within a population (De Cáceres et al. 2011).
We used Bayesian general mixed-effects models to estimate the repeatability of food preferences, using the packages MCMCglmm (Hadfield 2009) and BRMS (Bürkner 2017). For both wild-trapped adults and captive-bred juveniles, we partitioned the within- vs. between-individual components of variance in food preferences by fitting individual identity as a random effect. The among-individual variance expressed as a proportion of the trait is the repeatability (Gamer et al., 2010). The confounding effect of sex, age, session, and aviary were evaluated by including them as fixed effects. Although individuals were tested in groups of six individuals, we did not include the testing group as random blocking effect because adding it did not improve model fit (Table S1). To assess whether morphology influenced food preferences of pigeons, we conducted a Principal Component Analysis (PCA) based on the correlation matrix of morphological traits with the function “prcomp” in R (R core team 2015), and used the scores as predictors in mixed-effects models.
We disentangled the genetic and environmental components underpinning variation in food preference using an animal model approach (Kruuk & Hadfield 2007; Wilson et al. 2010), again based on Bayesian mixed-effects models. For labile traits, phenotypic variation at the between-individual level is not necessarily equal to additive genetic variation because individuals may also experience differing sets of non-genetic effects on their phenotypes. However, the between-individual phenotypic variation can be decomposed into genetic and non-genetic (or so-called ‘permanent environmental’) components using quantitative genetic models (Kruuk & Hadfield 2007; Wilson et al. 2010). Thus, we used Bayesian Mixed Models that included individual identity plus two additional variables as random factors, one relating individuals to their records in the pedigree and another accounting for the nest where offspring were raised (as half offspring were cross-fostered). We estimated heritability by dividing the posterior distribution of the variance component associated with the pedigree by total variance.
We also used the “animal model” to analyse diet plasticity by means of reaction norms, comparing preferences during the short-term assays with those from the long-term assays.
Following Nussey et al. (2007), we assessed variation in plasticity of the niche centre among individuals by including the interaction between period (i.e. short-term vs long-term, coded as a fixed effect) and the identity of individuals (coded as a random effect). We compared this model with a model including the pedigree as a random effect to determine the existence of heritable variation in plasticity (i.e. adaptive plasticity). We compared models using Waic, the widely applicable information criterion (Bürkner 2017).