Statistical Analysis
We analyzed community alpha diversity using log (observed ASV richness) and the Shannon Diversity Index. For modeling diversity, we used a linear mixed model as implemented in the R package lme4 (Bateset al. 2007) and evaluated the importance of different variables, taking into account the repeated sampling of some birds. We included host age (SY or ASY), sex (male or female), year (2017 or 2018) and sampling period (The Bahamas, first recapture in Michigan, and second recapture in Michigan) as fixed effects and individual host as a random effect. Using lmerTest (Kuznetsova et al. 2015), we generated an ANOVA table from the linear model analysis, and subsequently conducted a posteriori pairwise tests to compare the three sampling periods. Additionally, we conducted a pairwise t-test to assess differences between tagged and randomly caught birds within the first recapture period of 2018. We tested for the influence of outliers, which appeared to cause a deviation from normality in ASV richness (Shapiro-Wilks test), by repeating the analyses with outliers omitted and obtained very similar results. Finally, we tested for the effect of individual-level random effects with a likelihood ratio test comparing the model with and without individual ID as the random effect term, and we found individuals did not consistently differ from each other.
To examine community differences in the microbiome (beta diversity), we applied permutational multivariate analysis of variance (PERMANOVA) of Bray-Curtis dissimilarity and unweighted UniFrac distances, calculated among individual samples (Anderson 2014). For variables that showed significant differences in the PERMANOVA analyses, we conducted ana posteriori test to assess differences in dispersion or centroids using PERMDISP. We visualized beta diversity of significant variables using non-metric multidimensional scaling (nMDS) ordination of the Bray-Curtis measurements. Diversity calculations were implemented using the R packages vegan and phyloseq (Oksanen et al. 2007; McMurdie and Holmes 2013). Finally, to ask which taxa differ in abundance across sampling periods, we implemented analysis of composition of microbes (ANCOM) in QIIME2 (Mandal et al. 2105). ANCOM utilizes the underlying structure of the microbiome data to identify differentially abundant taxa between categories.