Testing for differentiation of microbial communities using phylogenetic methods: accounting for uncertainty of phylogenetic inference and character state mapping.

Comparative analyses of microbial communities increasingly involve the assay of 16S rRNA (or other gene) sequences from environmental DNA. Determining whether the composition of two or more communities differ in their phylogenetic composition involves testing for covariation between phylogeny and community type. This approach requires estimating the phylogenetic relationships among all sampled sequences and assessing whether the distribution of sequences among communities differs from the null expectation that sequences are randomly distributed. One method developed for implementing the phylogeny-based test of differentiation, referred to as the Phylogenetic test, relies on a single estimate of the phylogeny. However, for most data sets, many alternative phylogenetic trees provide statistically equivalent descriptions of the data. Because the actual phylogeny is unknown, phylogenetic tests of differentiation among microbial communities must account for phylogenetic uncertainty. In this article, we evaluate bootstrapping and Bayesian phylogenetic methods when implementing the Phylogenetic test using parsimony to map character states, and we investigate the effects of character mapping uncertainty by using a Bayesian approach to stochastically map character states on trees. Our approaches incorporate uncertainty into the tests of two closely related null hypotheses: (1) populations are panmictic, and (2) identical communities existed in both environments over the course of evolutionary history. We use two data sets previously implemented in tests for community differentiation: nitrite reductase genes sampled from marsh and upland soils and 16S rDNA sequences sampled from the human mouth and gut. We show that accounting for phylogenetic and mapping uncertainties can drastically affect results when implementing the Phylogenetic test. Accounting for phylogenetic and character mapping uncertainty provides a more conservative and robust test of covariation between phylogeny and environment when comparing microbial communities using DNA sequences.