Charting uncertainty about ant origins.

O ver a wide range of environments, up to five ant species forage every square meter of ground (1). In Amazonian rainforests, the biomass of ants dwarfs that of vertebrates (2), and in many rainforest trees, ants make up a large fraction of individual insects (3). This ecological dominance and the complexity of their societies makes their phylogeny of great interest as a glimpse into the development of the modern world in terms of the relationships between the various groups of ants, how their characteristics evolved, and when they originated. This year we have seen not one but two blockbuster articles examining ant phylogeny and time of origin of the group, one of which is by Brady et al. (4) in this issue of PNAS. The two articles (4, 5) agree in several important respects but disagree in others. Early thought on ant phylogeny was bedeviled by the belief that all or most of the genera with armored cuticles and strong stings belonged in a single subfamily, the Ponerinae (6). Brown (7) pointed the way forward by suggesting that various other ant subfamilies arose within the ponerines, which are thus paraphyletic; presciently, he proposed a close relationship between the Ectatomminae (then a ponerine tribe) and the giant subfamily Myrmicinae [ 4,500 species (8)]. However, he made no nomenclatural change, and subsequent authors tended to treat the ponerines as a single group. This tendency to agglomerate seriously compromised the ability to make sense of ant phylogeny, and for decades the procession of phylogenetic schemes was notable in its diversity rather than its stability. The crucial breakthrough came from Bolton (24), who erected a host of new subfamilies and subdivided the original subfamily Ponerinae into six; although he still placed all of these together, this recognition of difference liberated phylogeneticists to make new findings (4, 5, 9). Bolton’s (24) reorganization of ant systematics joined with the increasing ease of obtaining DNA sequence, a moderately good fossil record, and the rise of phylogenetic methods able to handle large data sets and estimate divergence dates. The first major and convincing effort to elucidate ant phylogeny at a grand scale and set it in temporal context was that of Moreau et al. (5) earlier this year. The study by Brady et al. (4) is even larger, dealing with 162 species from all 20 currently recognized ant subfamilies and 10 outgroups and using 6 kb of DNA sequence from seven nuclear genes. There is much agreement between the two studies. In particular, most subfamilies are monophyletic, and the two trees place them in similar positions. Brown’s suggestion of a strong relationship between the Ectatomminae and the Myrmicinae is not contradicted statistically by the new findings. There is thus now the emergence of the promise of stability in ant phylogeny, with these studies having very similar trees, but this result includes a puzzling anomaly, namely the placement of the Leptanillinae as the sister group to all other ants. Those leptanillines that have been studied are tiny, eyeless subterranean ants with an army-ant lifestyle, preying on geophilomorph centipedes like wolves on elk (10). Their bizarre habit of the queen feeding on hemolymph from her larvae also occurs in the Amblyoponinae (11), and this and morphological similarities raised suspicions that these groups are closely related. Having the Leptanillinae placed at the base of the tree of all ants (4, 5) is therefore very odd. For one thing, eyeballing the resulting tree gives the impression that the ancestral ant was eyeless and lived underground, so that the great majority of ants today must have secondarily regained eyes and moved to hunt in the open air. Long branch attraction (12), in which groups at the ends of long branches are wrongly placed together during phylogeny inference, can also lead to spurious rearrangement of the ingroup taxa (13). The problem is mainly one for parsimony and will not occur for maximum likelihood or Bayesian analysis when the substitution model has been correctly specified (14), but the models now available may not reflect reality sufficiently well to avoid it in some small, but unknown, number of cases (15). To paraphrase Li (16), substitution models are naturally artificial despite the attempt to be artificially natural. Brady et al. (4) surmised that long branch attraction might have affected the placement of the ant groups and thus repeated the analysis with the outgroups omitted. Significant differences appeared between the two analyses (Fig. 1). In particular, the poneroids, a group of morphologically similar subfamilies, which had formed a monophyletic assemblage in the rooted tree, no longer did so when the outgroups were omitted. Next, Brady et al. tested nine hypotheses for the rooting of the ant tree by constraining each such link in turn and found that only four of these were eliminated statistically. The most likely one of the remaining five still placed the Leptanillinae as the sister group to the rest of