A reticulate evolutionary history of the three subspecies
To infer the lineage-splitting pattern within A. marina , we fitted several models using approximate Bayesian computation (ABC) to test whether we can distinguish population histories. Our ABC approach shows that simulated sequences under the model with each subspecies diverging simultaneously provides the best fit to the observed data (Figure 5a). This conclusion was validated by three repetitions and high posterior probability of this model (> 0.6, Table 2). This result indicates the three subspecies diverged from each other simultaneously. In contrast, simulations with artificial groups (Simulation 2) allow no robust model selection.
The BB population morphologically diagnosed as marina shows lower genetic divergence and differentiation from eucalyptifolia thanmarina (Figure 2). Is it an eucalyptifolia mis-diagnosed as marina or a marina exchanging genes witheucalyptifolia ? Our ABC simulation (Simulation 3) shows that BB has descended from marina but experiences gene flow witheucalyptifolia populations (model v2, posterior probability 0.933, Table 2 and Figure 5b). This indicates that subspecies, while significantly differentiated, are genetically permeable. We also used TreeMix to capture potential gene flow events among populations. We identified five such events on the population splitting graph (Table S7). Three such events occurred between subspecies and the other two events occurred between BB and the outgroup species A. alba(Figure 5c).