Is Wing Recurrence really impossible?: a reply to Trueman et al .

By using multiple molecular markers and employing several methods of tree reconstruction and character optimization, we demonstrated that the ancestral phasmid is reconstructed unambiguously as wingless, with wings being reacquired later in phasmid evolution (Whiting et al., 2003). We presented this as a compelling example of recurrence in which a complex character, once lost to evolution, is regained subsequently in a descendant lineage (West-Eberhard, 2003). Our hypothesis is refutable via additional phylogenetic analyses including a larger selection of taxa, additional molecular markers, morphological data, or by examining patterns of development of wing expression in phasmids. We are currently performing research in each of these areas to add greater precision to this hypothesis. Trueman et al. (2004) have not presented a formal test of our hypothesis, nor contributed additional data to refute our findings. Wing recurrence is a hypothesis of character transformation, requiring a phylogenetic topology for interpretation. The ‘traditional view’ of phasmid wing evolution that these authors embrace was conjured in phylogenetic ignorance, since we presented the first formal analysis of phasmid phylogeny. Clearly the current data support a basal placement of apterous taxa, and multiple researchers who have reanalysed our data, including Trueman et al., have been unable to find a topology which rejects this hypothesis, regardless of analytical methodology. Thus, Trueman et al. quibble over methods of character optimization by launching into confused, non-phylogenetic, and mutually contradictory arguments to unravel our hypothesis ‘before this extraordinary evolutionary scenario reaches the entomology textbooks’. When the cost of wing gain is set extremely high (parsimony), or the rate of transformation from wingless to winged is set extremely low (likelihood), any method of character optimization will bias against reconstructing wing recurrence. More generally, values always can be selected to make it impossible to detect character recurrence by forbidding its transformation on a phylogenetic topology (1⁄4 Dollo’s law). Because we believe that phylogenetic topologies should establish patterns for inferring evolutionary processes, the issue becomes how much evidence is required before recurrence is a well-supported hypothesis. Our analyses and those of Trueman et al. agree on one critical point: under both parsimony and likelihood methods of character optimization the ancestral stick insect is supported unambiguously as wingless, with wings gained on multiple occasions. Trueman et al. (2004) premise their argument with the curious statement that ‘reconstruction of the phasmid ancestor is not the relevant issue.’ We argue it is the only relevant issue. If an ancestral node is reconstructed as wingless, and a descendant node is reconstructed as winged, then there must be a transformation from wingless to winged. Gains and losses are not observations: they are inferences based on hypotheses of character transformations given a topology and method of character optimization. To discard ancestral reconstruction is to discard the very reason why phylogeny is critical for investigating character evolution. Trueman et al. fail to follow their own dictum in discussing the relative merits of hypotheses using values obtained directly from character optimization (e.g. four gains and three losses for the wing ‘re-evolved’ hypothesis). Moreover, if all that matters is how well the ‘model fits the data’ then we can dispense with their parsimony arguments outright, as most systematists would acknowledge that parsimony is not attempting to quantify the fit of data to specific evolutionary models. Trueman et al.’s criticisms can be summarized as three points: (1) the ‘probability’ of wing loss to gain should be 2.5 under parsimony; (2) the ‘probability’ of wing loss to gain should be 6 : 1 under parsimony; and (3) the ‘probability’ of wing loss to gain should be 13 : 1 given a likelihood analysis.