In a study recently published in BMC Evolutionary Biology, Oliver and Prudic [9] now revisit the case. They use sequence information from eight nuclear loci combined with coalescent simulation of gene trees to evaluate a range of models of population structure and evolutionary history. By using multiple loci they compensate, at least partially, for the problem of gene-tree/species-tree discrepancies. The main difference from earlier approaches, however, is that Oliver and Prudic use sophisticated statistical models to measure how well their simulations fit the empirical data. Parameters for these 15 models were estimated divergence times (for species evolution) and migration rates (for population structure), taken from earlier studies of the same species. The advantage of this approach is that explicit models for contrasting evolutionary scenarios are compared with each other. Hence, inference is based on rigorous statistical tests of explicitly formulated alternatives. Oliver and Prudic found that the MMH had to be rejected: the only model that fitted the data in all aspects was a scenario that assumes moderate migration rate of the butterflies plus divergence times of about 655,000 years for the split of arthemis from astyanax, and 1,075,000 years for the split of the western arizonensis from the eastern (arthemis + astyanax) clade.

Is this the end of the story? Certainly not. First, even if monophyly of the two mimetic forms now seems to be rejected with good support, this is not yet firm evidence for a reversion of the ancestral phenotype from the mimetic one. Two independent gains of Batesian mimicry in astyanax and arizonensis could still have occurred, while arthemis just retained the plesiomorphic character state. This scenario would require two steps in character evolution - exactly the same number as one gain of mimicry at the base of the arthemis complex, and one loss subsequently at the split between arthemis and astyanax. To decide conclusively between these competing scenarios a better understanding of the genetic basis and physiological processes that determine the two different phenotypes in the L. arthemis group will be required. While prima facie butterfly wing patterns might be seen as complex characters, with a low likelihood of convergent evolution, in fact most cases of butterfly mimicry are based on increased melanism. Major 'macro-evolutionary' changes in wing color patterns could thus be controlled by very few genes, or even one single major developmental gene [10, 11]. In that case, convergent evolution of similar melanic (and at the same time mimetic) phenotypes remains a plausible alternative. One obvious approach to a better understanding is therefore to unravel the developmental pathways that lead to mimetic phenotypes and their genetic basis, by genomic analysis, for example [11].

Second, the phylogenetic and statistical analyses of Oliver and Prudic are not immune to criticism. The sample sizes for some genes were very small. To obtain a more comprehensive picture of the history of Limenitis phenotypes in North America, including possibly complex patterns of gene flow, a thorough phylogeographic study would be required, using a larger number of populations from the entire range of the complex, a larger number of genetic markers, and incorporating the allied species L. weidemeyerii and lorquini with a larger number of samples. That last requirement seems important, as hybrids between arthemis and these two relatives do occur. If introgression between species and subspecies has been a significant phenomenon in the phylogeographic history of North American white admirals, it will have left traces in the genetic architecture as well as in the phenotypes of L. arthemis - which would be likely to go unnoticed in too small samples.

This new study on the evolution of mimicry in butterflies exemplifies the fact that, even in putatively well-studied cases, many questions about evolutionary processes remain to be settled. The application of phylogenetic hypothesis testing allows a great step forward as it provides measures of support that can be compared across competing scenarios. As with all statistical models, the problem of parameterization remains. Even for parameters such as divergence times and migration rates, empirical estimates are often unrealistic - for example, in the absence of fossil evidence for calibration or when information on a species' population biology is scant. Nevertheless, with increasing availability of computing facilities and pertinent software the lesson to be learned is that explicit comparison across competing phylogenetic hypotheses is now one, among many, approaches to unraveling the evolution and function of the fascinating diversity of butterfly wing patterns.