Extinction and the Loss of Evolutionary History
Phylogenetic Structure in Extinction Risks
We have discussed above how the process of extinction is non-random with respect to species traits and geography. For example, extinction will tend to remove largebodied species with slow life histories and narrow niches, and species in regions with high intensity of extinction drivers. Because many of the traits linked to extinction risk (e.g. body size, generation time, dispersal ability etc.) demonstrate phylogenetic conservatism (Fritz and Purvis 2010), such that they tend to be clustered on the phylogeny, extinctions will also tend to cluster on the phylogeny. Whereas evidence for trait-based explanations for plant extinctions is mixed (Freville et al. 2007; Bradshaw et al. 2008; Sodhi et al. 2008; Davies et al. 2011; Daru et al. 2013), phylogenetic selectivity in extinction risk might also result from a geographical pattern in the drivers of extinction, for example, range elevation might determine a species vulnerability to climate change (Sandel et al. 2011). If closely related species also tend to have close geographical proximities, perhaps reflecting shared habitat preferences or the geographical process of speciation, they will then also be exposed to similar intensity of extinction drivers. There is an increasing weight of evidence suggesting that extinction risk is generally more clustered on a phylogeny than expected by chance (Bennett and Owens 1997; Purvis et al. 2000a; Schwartz and Simberloff 2001), a pattern also observed within the fossil record. Extinction will thus prune the tree-of-life non-randomly. However, how this non-random pruning might impact the loss of evolutionary history has been a subject of recent debate.
Quantifying the Loss of Evolutionary History
Extinction prunes species from the tips of the tree-of-life, resulting in the loss of terminal branches. In a frequently cited paper, Nee and May (1997) used simulations to explore the expected loss of evolutionary history (quantified as the summed branch lengths from the tree-of-life) under various extinction intensities. Perhaps surprisingly, they found that up to 80 % of the tree would remain under even extreme extinction scenarios in which 95 % of species were lost. However, their simulations were unrealistic in two regards. First, they assumed extinction events were random – the field-of-bullets model, in which extinction is independent of species' traits and thus also phylogeny. If extinctions are clustered on a phylogeny, we might also lose the internal branches of the tree that connect them, and thus experience a greater overall loss of phylogenetic diversity (Russell et al. 1998; Purvis et al. 2000a). Second, their expectation was derived assuming a phylogeny based on a coalescent model, which generates a highly unrealistic distribution of branching times, with most branches clustered towards the present (see Fig. 1a), and does not fit to most empirical estimates of phylogenies. Importantly, coalescent trees tend to be 'tip-heavy' such that most branching events are short and clustered towards the present (tips of the tree). Therefore, under this model, most extinctions remove only short terminal branches from the tree, and most major lineages survive even extreme pruning of tips. Empirical phylogenies tend to have a very different distribution of branching times (e.g. Rabosky and Lovette 2008; see also Fig. 1b, c for pure birth and birth-death tree). Mooers et al. (2012) explore further how tree shape impacts the expected loss of phylogenetic diversity. The phylogenetic non-random distribution of extinction risk and the shape of empirical phylogenies might therefore suggest that we risk losing a disproportionate amount of evolutionary history from the tree-of-life.
A suite of empirical studies were to follow on from the early work of Nee and May, and emphasized both the phylogenetically non-random nature of species' extinctions and a greater than random loss of phylogenetic diversity (e.g. Purvis et al. 2000a; Purvis 2008; Vamosi and Wilson 2008). A link between non-random
Fig. 1 Comparison of branching times for different tree reconstruction models of size 128 tips. a Coalescent model in which branching clusters towards present; b pure birth model in which all lineages have an equal probability of splitting (b = 1.0) and no lineages go extinct (d = 0); c birthdeath model in which lineages have equal rates of splitting and extinction (birth = 1.0, death = 0.2)
extinction and greater than random loss of phylogenetic diversity seemed intuitive; if two sister species are lost to extinction, not only do we lose the unique phylogenetic diversity captured in the branches from which they subtend, but we also lose the ancestral branch that is shared between them (see Fig. 2). However, in a more recent study, again using simulations, but this time assuming both a more realistic model of diversification and a range of phylogenetic signal in extinction probabilities, Parhar and Mooers (2011) suggested that the loss of phylogenetic diversity under phylogenetically non-random extinctions was more or less indistinguishable from random (see also Heard and Mooers 2000). Seemingly, the observation of phylogenetic signal in extinction risks and the non-random loss of phylogenetic diversity are not necessarily connected directly.
Observations for greater than random losses of phylogenetic diversity that have been inferred for many clades under realistic extinction scenarios likely reflect the
Fig. 2 Ultrametric phylogenetic tree with three tips (A, B and C) and four branches with lengths in millions of years (Myrs). If tip taxa A and C become extinct, we lose two branches and 3 Myrs of evolutionary history from the tree. If sister taxa A and B become extinct, for example, because they share a phylogenetically conserved trait that predisposes them to high risk, we also lose 3 Myrs of evolutionary history, but this time three branches are lost from the phylogeny
particularities of phylogenetic tree topology in combination with a tendency for more extinction prone species to fall within species poor clades (Heard and Mooers 2000; von Euler 2001; Parhar and Mooers 2011). There does seem to be a general trend within some clades for threatened species to be overrepresented in speciespoor clades (e.g. in mammals, Purvis et al. 2000b and birds, Bennett and Owens 1997). In plants, patterns appear mixed. As discussed above, there is some evidence suggesting an opposite trend to vertebrates, with a greater proportion of threatened plant species falling within species-rich clades (Schwartz and Simberloff 2001; Lozano and Schwartz 2005), and less evolutionary distinct lineages (Davies et al. 2011). Globally, however, species poor, and especially monotypic plant families, again appear to be more threatened, and their extinction would also result in a disproportionate loss of evolutionary history (Vamosi and Wilson 2008).