Temporal changes in telomere and tarsus lengths during artificial selection
In order to analyze how nestling tarsus length and TL were affected by the artificial selection for longer (high ) and shorter tarsi (low ) during the study from 2002-2006, we used linear mixed effects models (R package lme4 , Bates et al. 2015) including year (i.e. birth cohort 1 to 5) as a continuous predictor variable, as well as the quadratic effect of year (year2). Tarsus length and TL are expected to change during development within individual nestlings (Hall et al., 2004; Boonekamp et al., 2014) and there might be sexual differences in morphology (Cordero et al., 2000) and telomere dynamics (Barrett & Richardson, 2011). Thus, nestling age (number of days since hatching) and sex were included as explanatory variables in all models. Selection status category (0, 0.5, or 1) was included in addition to an interaction term between selection status and year. All models assumed a Gaussian error distribution and included a random intercept for brood identity to account for the non-independence of nestlings from the same brood. We structured these analyses into four sections, where we analyzed each selection regime (high or low population) separately for each response variable (tarsus length and TL). In order to identify the predictors most supported by the empirical data we constructed and compared alternative candidate models (Burnham & Anderson, 2002) fitted with maximum likelihood within each section using Akaike’s information criterion (Akaike, 1973) corrected for small sample sizes (AICc, Hurvich & Tsai, 1989). All models were validated visually by diagnostic plots and model parameters are given from models refitted with restricted maximum likelihood (REML). To reduce the problem of multicollinearity in multiple regression analyses, we only included predictor variables with intercorrelation Pearson’sr <0.5 for all relevant pairs of explanatory variables. All statistical analyses were performed in R version 3.5.2 (R Core Team, 2018).