Demographic histories can also be estimated from genealogies (phylogenies) in a Bayesian statistical framework using BEAST (versions 1.4.6 and 1.7.2, Drummond and Rambaut 2007; http://beast.bio.ed.ac.uk). MODELTEST (Posada and Crandall 1998), using the Akaike information criterion, indicated that the nucleotide substitution model of Hasegawa et al. (1985) was appropriate Natural Product Library cost when additionally allowing for unequal substitution rates among sites and for a proportion of sites to be invariable. When using BEAST, runs were of sufficient length

(typically 30 million or more) that effective sample sizes (ESSs) were always over 100, and usually very far over this value. Several runs were done for each input file to check for convergence. The program TRACER v1.4 (http://beast.bio.ed.ac.uk/) was used to analyze the output from BEAST. The first 10% of iterations in each run were discarded as burn-in. A coalescent exponential expansion model was specified and a randomWalkOperator selected for the exponential.growthRate parameter (Supplementary data files 2 and 3). If the 95% highest posterior density (HPD) of the growth rate parameter check details includes zero, a hypothesis of constant population size cannot be rejected (Marino et al. 2011; https://groups.google.com/d/msg/beast-users/y-ppM_dB5UI/uPybHlRMYc4J).

Monophyly of Australian dugongs was forced, and a prior of 115,000 yr (normal distribution ± 5,000) (date of the closure of Torres Strait to transit by marine organisms at the start of the last glacial period inferred from sea-level estimates; Fig. 2)

placed on the most Cyclin-dependent kinase 3 recent common ancestor (MRCA) of all Australian dugongs (see Supplementary data file 6). Trees generated during this analysis were examined for the strength of support given to the individual lineages. Bayesian skyline plots (BSPs) (Drummond et al. 2005) show changes in effective population size (NE(FEMALE)) over time, along with credibility intervals. A major advantage of this approach is that it avoids problems associated with choosing a single demographic scenario such as “constant population size” or “exponential growth.” Sample input files are in Supplementary data files 4 and 5. The mutation rate prior was specified (following the analysis above) as normally distributed with a mean of 24.8% per million years and lower and upper bounds of 13.89% and 37.46% per million years, respectively. Given that most samples came from distinct localities where sampling was possible, we simply used those localities as the basis for assigning individuals to “populations.” With some clustering of localities by geographic region if samples were few in number, we could define 11 populations on this basis (each represented by a pie chart in Fig. 1). There are no clear criteria for a priori grouping of these populations for a hierarchical analysis such as AMOVA (Excoffier et al. 1992).