The author’s formulation is conceptually simple (Fig. 1). One first defines evolutionary history as the sum of all the branch lengths connecting the species in a rooted phylogenetic tree (more on that below). This history is then partitioned across a study landscape that has been divided into equal-sized grid cells. In the simplest case that the authors present, every branch on the tree is sliced into equal-sized pieces, with the number of pieces being decided by how many grid cells the species (or members of the larger group the branch leads to) are found. Once all branches are divided up, the slices are distributed additively onto the relevant grid cells.
In this way, cells with many species, cells containing species with long unshared branches on the paths back to the root, and cells that contain species that are found in fewer other places get more and larger slices of the branches. Cells that contain species that are both evolutionarily distinctive (i.e. have few close relatives) and that are also not found elsewhere are awarded the highest amount of this combined ‘phylogenetic endemism’ or PE. At the geopolitical scale, New Zealand would probably contain inordinate amounts of PE, given the archaic frogs, the tuatara, and the evolutionarily isolated riflebird, kea and kakapo found exclusively there.
Rosauer et al. make formal connections between PE and two well-characterized measures, the weighted endemism (WE) and the total phylogenetic diversity (PD) found in a particular cell. In theory, since all three are sums across species, all three metrics should scale with simple species richness (SR). However, for the two clades the authors use to illustrate their method (Figs 2 and 3), PE correlates less with SR than does either WE or PD (for Davesia, SR vs. PE, WE and PD: r = 0.52, 0.77, and 0.74, respectively; for Pelodryadinae, r = 0.36, 0.54, and 0.98; D. Rosauer, personal communication). For Pelodryadinae frogs and Davesia, this decoupling allows the top ranking areas on the WE scale to be different from the top ranking areas using the combined PE scale. If this is generalizable, attending to areas with the most small-range species would not effectively conserve the tree of life.
More modelling and comparative work is needed to elucidate why, and to what extent, measures of endemism and PD diverge from measures of PE. One possibility is that phylogenetically isolated species are often range restricted. If true, given that phylogenetically isolated species are rare in most trees, the majority of grid cells would be populated entirely by species from ‘bushy’ parts of the phylogeny. This would mean that the variation among these cells in, for instance, PD value, would be due to loss or gain of short terminal branches, and as a result, PD value would be primarily a function of SR (as it is for the frogs here). Across the whole landscape, the PE measure instead strongly weights the distribution of the few phylogenetically isolated species (e.g. Davesia microcarpa, Fig. 2) and it would be their spatial distribution that drives the differences in the correlations coefficients seen here.
Besides evolutionary and ecological mechanisms, there are other pressing issues. If phylogenies are to become of practical use to conservation managers, then the ‘total evolutionary history’ that they are supposed to represent must be clearly defined, most usefully in terms of feature diversity (Faith 1994). A cursory look at the trees inferred by Rosauer et al. reveals the common situation. Far from the idealized bifurcating ultrametric tree in our Fig. 1, these are ragged, non-ultrametric, incomplete consensus gene trees with many polytomies. So what exactly are we slicing when we slice up branches on such a tree?
Rosauer et al. are slicing up the inferred number of molecular substitutions in a few gene segments. If we are to interpret the results in a conservation context, this inferred number of substitutions must be related to conservation-relevant variation: we know of no general tests of such a relationship. One can bypass a DNA substitution – genetic variation – conservation-relevant phenotypic variation connection by appealing to phylograms: Mooers et al. (2005) argued that time might be the most general and fungible metric of PD, and Rosauer et al. do suggest that testing the robustness of their results using time-based versions of their trees would be useful. However, to the extent that conservation-relevant diversity does not diverge in a time-dependent fashion (see also Williams et al. 1994), and to the extent that the ultrametric trees (like that depicted in our Fig. 1) needed for time-based analyses are most often just massaged gene trees (with the molecular clocks carefully ‘rate-smoothed’ or ‘relaxed’), that argument may be less sophisticated than simplistic. More work is definitely needed in this area.
Species that have a known distribution on the landscape but are not yet placed on the phylogeny must also be treated fairly. Rosauer et al. take a justifiably pragmatic approach, attaching species to the tree using taxonomic information, but assigning little unique branch length to them. This does penalize them, though, since they contribute less on average to the cells where they are found. Another approach might be to model their branch lengths using ‘local’ phylogenetic information (e.g. giving them the mean branch length of their congeners). If time is what branch lengths are to represent, birth–death models of evolution could be employed in conjunction with genetic data to generate a set of reasonable histories. This set also need not be summarized by consensus trees prior to use, avoiding additional problems associated with polytomies (which often suggest more unique evolution than is true).
Geographically rare and phylogenetically rare species both need special care, and we need to know where they are to most effectively give them that care. Rosauer et al. have moved us forward with a sophisticated view of where the tree of life is geographically concentrated, maybe even highlighting those areas that deserve particular conservation attention. We need further refinements, perhaps even simplifications of their approach, including ways to combine measures across taxa; studies that investigate how the two types of rareness are related on the landscape; and, most pressingly, implementations of such rational, quantitative approaches in active conservation management.