In all of the diversification models that we have considered so far, speciation happens instantly; one moment we have a single species, and then immediately two. But this is not biologically plausible. Speciation takes time, as evidenced by the increasing numbers of partially distinct populations that biologists have identified in the natural world (Coyne and Orr 2004; De Queiroz 2005). Furthermore, the fact that speciation takes time can have a profound impact on the shapes of phylogenetic trees (Losos and Adler 1995). Because of this, it is worth considering diversification models that explicitly account for the fact that the process of speciation has a beginning and an end.
The most successful models to tackle this question have been models of protracted speciation (Rosindell et al. 2010; Etienne and Rosindell 2012; Lambert et al. 2015). One way to set up such a model is to state that speciation begins by the formation of an incipient species at some rate λ1. This represents a “partial” species; one can imagine, for example, that this is a population that has split off from the main range of the species, but has not yet evolved full reproductive isolation. The incipient species only becomes a “full” species if it completes speciation, which occurs at a rate λ2. This represents the rate at which an incipient species evolves full species status (Figure 12.5).
Because speciation takes time, the main impact of this model is that we predict fewer very young species in our tree – that is, the nodes closest to the tips of the tree are not as young as they would be compared to pure-birth or birth-death models without protracted speciation (Figure 12.6). As a result, protracted speciation models produce lineage through time plots that can mimic the properties often attributed to diversity-dependence, even without any interactions among lineages (Etienne and Rosindell 2012)!
Likelihood approaches are available for this model of protracted speciation. Again, the likelihood must be calculated using numerical methods (Lambert et al. 2015). Fitting this model to the salamander tree, we obtain a maximum log-likeihood of 513.8 with parameter values λ1 = 0.059, λ2 = 0.44, and μ = 0.0. This corresponds to an AIC score of -1021.6; this model fits about as well as the best of the time-varying models but not as well as the diversity dependent model considered above. Again, though, I am not including plausible combinations of models, such as protracted speciation that varies through time.
So far, models of protracted speciation remain mostly in the realm of ecological neutral theory, and are just beginning to move into phylogenetics and evolutionary biology (see, e.g., Sukumaran and Lacey Knowles 2017). However, I think models that treat speciation as a process that takes time – rather than something instantaneous – will be an important addition to our macroevolutionary toolbox in the future.