Special Issue: Ecological and evolutionary informatics Space, time, form: viewing the Tree of Life Roderic D.M. Page Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow, G12 8QQ, UK There are numerous ways to display a phylogenetic tree, which is reflected in the diversity of software tools available to phylogenetists. Displaying very large trees continues to be a challenge, made ever harder as in- creasing computing power enables researchers to con- struct ever-larger trees. At the same time, computing technology is enabling novel visualisations, ranging from geophylogenies embedded on digital globes to touch-screen interfaces that enable greater interaction with evolutionary trees. In this review, I survey recent developments in phylogenetic visualisation, highlight- ing successful (and less successful) approaches and sketching some future directions. Visualising trees Visualising phylogenies is one of the fundamental tasks of evolutionary analysis. Reviews of the field [1,2] list a growing number of tree viewers, some of which, such as NJPlot [3] and TreeView [4], have been in use for over a decade. A quick glance at Felsenstein’s list of phylogeny programs (http://evolution.genetics.washington.edu/phylip/ software.html#Plotting) reveals viewers for just about every conceivable operating system, written in a wide range of computer programming languages. Given this diversity of tools that all provide essentially the same functionality, it would be tempting to conclude that the basic problem of displaying an evolutionary tree has been solved. Yet, it was striking that all the entries in the iEvoBio 2010 visualisation challenge were tree viewers (Figure 1). This suggests that although the niche of tree viewer is crowded, biologists working with trees are still searching for tools to help them visualise phylogenies. The goal of this review is to survey some recent developments in phylogeny visualisation, with an eye to future directions. Trees are relatively simple structures that place few restrictions on how they can be depicted, apart from pre- serving the connections between the nodes in the tree. This lack of constraints has led to a proliferation of ways to visualise trees, many of which are striking (for a visual survey, see http://treevis.net). Conversely, this freedom means that the interpretation of a tree diagram might not always be obvious to the person viewing it [5] [Green, D. and Shapley, R. (2005) Teaching with a visual tree of life http://groups.ischool.berkeley.edu/TOL/], especially distinguishing which aspects of the diagram are providing information, and which largely reflect artistic license (Box 1). Although the most common representation of a phylog- eny is a two-dimensional (2D) Euclidean drawing [1], an increasingly diverse range of visualisations are emerging (Figure 2). Typically phylogenies are drawn as trees how- ever, authors have experimented with treemaps [6], which lay out a tree as a set of nested rectangles (Figure 2). Treemaps are perhaps best suited for classifications rather than phylogenies, although Arvelakis et al. [7] recently used treemaps to display phylogenies with over 2000 species. Euclidean geometry is reassuringly familiar, but it becomes difficult to accommodate very large trees within the confines of the printed page or a computer screen. One approach is to ‘fold’ or collapse nodes to save space (Figure 3). Several methods, such as degree of interest (DOI) trees [8], space trees [9] and expand-ahead browsers [10], exploit the natural hierarchy of rooted trees to com- press the tree into a smaller display area. The choice of which nodes in a tree to collapse can be made by the user, or the process can be automated [11,12] An alternative approach to saving space is to keep the tree unchanged, but instead distort the space in which the tree is being displayed, the best-known examples being hyperbolic viewers (Figure 2) [13,14]. Although capable of producing some stunning images (e.g. Figure 4a), these tools have gained little traction among users. In practice, users find them hard to navigate, and hyperbolic viewers in particular are best suited to classifications, which tend to be shallow (few nodes along the path from any tip to the base or root of the tree) and frequently have internal nodes of high degree (many immediate descendants). By contrast, a fully resolved phylogeny may be deep (in a tree with n leaves, there may be a path from leaf to root with n–1 nodes) and binary (each node having only two immediate descendants) consequently, phylogenies rarely look good in hyperbolic viewers. Some three-dimensional (3D) phylogeny viewers have forgone trying to truly display a phylogeny in three dimen- sions, and instead use the third dimension to provide a ‘fly through’ experience over a 2D tree, such as Paloverde [15] and the Wellcome Trust Tree of Life (http://www. wellcometreeoflife.org/). Although perhaps less disorien- tating than hyperbolic viewers, it is not clear that this provides a better way to navigate through a tree compared with a simple 2D visualisation. Although the case for 3D Review Corresponding author: Page, R.D.M. (Roderic.Page@glasgow.ac.uk) 0169-5347/$ – see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tree.2011.12.002 Trends in Ecology and Evolution, February 2012, Vol. 27, No. 2 113

tree viewers in general is not compelling, there are specific cases where it has been employed successfully, notably in geophylogenies. Geophylogenies The relationship between phylogeny and geography has long been a subject of intense interest. Pioneering work by Hennig [16] and Brundin [17] became further developed into cladistic biogeography, which emerged at the same time as somewhat polarised debates about the relative roles of vicariance and dispersal in biogeography. The vicariance versus dispersal debate largely took place in the macroevo- lutionary arena by contrast, at the microevolutionary scale, extensive genetic and geographic sampling at the popula- tion level led to the rise of phylogeography [18], where displaying phylogenies on maps became widespread. Phylogeneticists have been relatively slow to make use of geographic information systems (GIS) tools [19], but the advent of Google Earth has inspired a flurry of activity [20,21]. Geophylogenies use latitude and longitude to lo- cate the leaves of a tree on a sphere representing the Earth, and use basic spherical geometry to position the internal nodes [22] (Figure 4b this is one instance where there is a plausible mapping of a phylogeny onto three dimensions). As with Euclidean phylogenies, the height of a node above the surface of the Earth in a geophylogeny can be arbitrary, or a function of the amount of genetic change, or time of divergence. The increasing availability of georeferencing services [23] and georeferenced DNA barcode sequences [24] means that geophylogenies are likely to become a much more widely used form of phylogeny visualisation [25]. The ability of Google Earth to display time series data can be used to combine spatial and temporal data to create compelling visualisations, such as the spread of zoonotic viruses [20,26,27]. At the same time, because the placement of nodes in the tree is constrained by geographic location, larger trees can quickly result in a mass of intersecting lines, which can sometimes obscure the visualisation. Relationships between trees At the outset of molecular systematics, pioneers such as the late Morris Goodman recognised that trees for genes need not match those for species [28]. Processes such as gene and genome duplication, gene loss and horizontal transfer can all create mismatches between gene and species trees. Population-level processes, such as incom- plete lineage sorting and introgression, can also produce incongruent gene and species trees [29]. Similar patterns of relationships among trees occur between phylogenies for hosts and their parasites, and in biogeography. In both cases, one entity is ‘tracking’ another over evolutionary time with greater or lesser fidelity [30]. Various methods have been proposed to view these interconnected trees. Some visualisations embed one tree inside another (Figure 5), a visualisation that is easy to construct [31] but that can quickly become visually noisy owing to the presence of multiple gene lineages in the same species. Kim and Lee [32] have explored using a 3D stacked layout (Figure 4c) to separate more clearly these otherwise over- lapping trees. PhyloBox (a) (b) (c) (d) (e) jsPhyloSVG GenGIS EOL tree viewer TRENDS in Ecology & Evolution Nexplorer Figure 1. Entries in the iEvoBio 2010 visualisation challenge. The five entries in the iEvoBio 2010 challenge were the interactive phylogeny viewers Phylobox (a) and jsPhyloSVG (b) [56], the GenGIS system for genomics data (c) [57], a treemap- based classification viewer from the Encyclopedia of Life (d) and the phylogeny and alignment viewer Nexplorer (e) [57]. Review Trends in Ecology and Evolution February 2012, Vol. 27, No. 2 114