2.6 | Diversity Analysis
The number of ASVs (ASV richness) observed within a given community was used as a measure of α-diversity. While α-diversity indicates within community diversity, β-diversity indicates differences in community composition between communities. We analyzed two β-diversity metrics: Euclidean distances from a centred log-ratio transformed ASV dataset (Gloor, Macklaim, Pawlowsky-Glahn, & Egozcue, 2017) and weighted UniFrac distance (Lozupone, Hamady, Kelley, & Knight, 2007) of a rarefied ASV dataset (34,280 reads/sample; rarefaction curve: Figure S1, Supporting information). Both β-diversity metrics weight differences in the relative abundance of bacterial taxa between communities, but weighted UniFrac measures simultaneously weight bacteria phylogenetic relatedness. Finally, we also considered β-dispersion, calculated as the distance from each sample to the sample-set centroid in Euclidean or weighted UniFrac space (Anderson, Ellingsen, & McArdle, 2006).
We evaluated the ability of spatiotemporal (Julian date, longitude and distance from the population’s midpoint), host physiology (using age and parental status as proxies), and habitat class relative area to predict patterns in the described microbiome diversity measures. Julian date, longitude, and longitudinal distance from the population’s centre were scaled to a mean of 0 and a standard deviation of 1 prior to analysis. The average longitude of all observations from the 2014 field season was used as a proxy for the population’s centre. Age was coded as continuous data in 1-year increments, with a linear and 2nd order polynomial fit considered in analyses, given a curvilinear relationship between gut microbiome diversity and age among humans (Yatsunenko et al., 2012). Parental status, shown to affect the microbiome in other systems (Amato et al., 2014), was coded as a dichotomous variable based on whether adult females were nursing a foal (<1 year old offspring) during the 2014 field survey.
For univariate diversity measures (α-diversity and β-dispersion), we used a multi-model inference approach implemented in the R package ‘MuMIn’ (Bartoń, 2009). A starting global general linear model was parameterized with the spatiotemporal, physiological, and environmental terms described above, without interactions. We determined parameter estimates and significance from conditional AICc averaging of models which had a Δ AICc < 3 (Burnham & Anderson, 2002; Grueber, Nakagawa, Laws, & Jamieson, 2011). Patterns in β-diversity were analyzed using a backwards selection approach from PERMANOVA outputs, with the global model outputs reported in the Supporting information (vegan R package, adonis2() function, parameter by = ‘margin’; Oksanen et al., 2019). Additionally, we ran a Mantel test to test for a correlation between spatial separation and β-diversity measures, and a univariate PERMANOVA to test for an effect of social band membership.