Portal de Periódicos da CAPES

Early Wagner trees and “the cladistic redux”

2012; Wiley; Volume: 28; Issue: 5 Linguagem: Inglês

10.1111/j.1096-0031.2012.00397.x

1096-0031

James S. Farris,

Genomics and Phylogenetic Studies

Sir, Hull (1988), Sneath (1995), Felsenstein (2004), Williams et al. (2010), and Williams and Ebach (In press) have all discussed the early history of quantitative phylogenetic methods, and their accounts are all misleading. Though differing in detail, their portrayals seem to reflect the common aim of obscuring the origin and significance of Wagner trees. To see this, first consider Hull’s (1988, p. 154) narrative: [In graduate school, starting in 1963] at Michigan Farris discovered that Herb Wagner had already published a method [groundplan-divergence analysis] for estimating phyletic relationships (Wagner, 1961), and Wagner was happy to work with Farris. Later, when Farris developed his own methods, he named his constructions “Wagner trees,” not “Sokal trees.” Arnold Kluge was also interested in the development of quantitative methods for reconstructing phylogenies. My methods at that time were based primarily on working out how to program Wagner’s method efficiently (a task that is simple today, but was less so then, as other Michigan students had learned). My efforts were first published (aside from my thesis) by Kluge and Farris (1969), and of particular interest here is that our treatment drew heavily on Wagner’s ideas. Our criteria for identifying ancestral states were Wagner’s,11 These could just as well have come from Hennig, but in practice I got them from Wagner. I already knew of Wagner’s approach in 1964, but naturally I did not read Hennig’s (1966) book until it was published. as was the now famous technique of constructing a tree by adding terminals one at a time in order of increasing advancement index. Also Wagner’s was the advancement index itself (a Manhattan metric), as well as the approach of reconstructing new hypothetical ancestors as terminals were added.22 Wagner loved to tell the story of reconstructing an ancestor in his thesis, then later finding just that combination of states in a newly discovered species. Moreover, we made the provenance of the method entirely clear. We did not say we had invented a new method that we were going to name after Wagner, but that it was the method of Wagner. That is why Hull avoided mentioning Kluge and Farris’ (1969) paper. Partial to Sokal, Hull felt that Wagner trees really should (for some undisclosed reason) have been named after Sokal. Readers could hardly be expected to sympathize with Hull’s sentiments, however, if Hull admitted that Wagner trees are actually based on W. H. Wagner’s approach, and so Hull simply pretended that connection away.33 It must be added that I had read a draft of Hull’s (1988) book—which draft, however, did not include the passage in question. The same applies, incidentally, to my (Farris, 1990) comments on Hull’s highly inventive coverage of the monophyly controversy. Sneath (1995, p. 288) went even further in that regard. His rendition included no one named Wagner, the “Wagner” in “Wagner trees” being left unexplained: The earliest numerical work on phylogenetic methods was by Edwards and Cavalli-Sforza (1964) and Camin and Sokal (1965)... It was found necessary to introduce into trees nodes that represent putative ancestors (Edwards and Cavalli-Sforza, 1964... This realization led to… minimum length trees and the mathematics of Steiner and Wagner trees. It is informative to consider that account in light of Fig. 1, which shows a Wagner tree (groundplan-divergence analysis) from Hardin (1957). Note that the hollow circles on that diagram denote reconstructed ancestors. But then Sneath (1995) did not really think that such ancestors (or trees) were unknown before 1964, as Sokal and Sneath (1963, p. 289) had already known better: Wagner tree (groundplan-divergence analysis), from fig. 15 of Hardin (1957). Hollow circles denote reconstructed ancestors, while letters below nodes identify derived character states. Attention should be drawn to Wagner’s method for expressing phylogenetic deductions, which permits quantification of the data. An example can be found in [Mickel (1962)]. After 1963, but before 1995, Wagner trees had played a role in criticisms of phenetic classification (see Farris, 1979). That is the likely cause of pheneticist Sneath’s (1995) apparent loss of memory, although his selective amnesia also had the benefit that Sneath’s friends turned out to have invented everything. Felsenstein (2004, p. 123) was also kind to his friends, and he came up with earlier work than Sneath had managed: In the original paper of Michener and Sokal (1957), the purpose of the clustering was not simply to classify, but to infer the phylogeny. There is thus a good case to be made that this was the first paper on numerical inference of phylogenies. Michener and Sokal’s “phylogenies”, however, were obtained by so interpreting the results of clustering by overall similarity, and that interpretation was soon retracted (Sokal and Sneath, 1963, pp. 229f): We have seen that one cannot derive evolutionary rates from similarity coefficients among recent forms... At the time they wrote their paper, Michener and Sokal (1957) were still attempting to establish phylogenetic classifications. The Wagner method, in contrast, actually is phylogenetic, based on the principle of grouping by shared derived states.44 For examples and further discussion of this point, see Farris (In press). Note, however, that it is Wagner trees that group by shared derived states. Unfortunately, Wagner’s own classifications included paraphyletic groups. That method, furthermore, was already well established when Michener and Sokal (1957) strangely decided to cluster by overall similarity instead (Hardin, 1957, p. 170): This method of expressing the phylogeny on a graph using concentric semicircles is one which is used by Dr. W. H. Wagner, Jr. in his classes on taxonomy at the University of Michigan. Like Hull and Sneath, then, Felsenstein arrived at his conclusion primarily by ignoring inconvenient parts of history. Williams and Ebach (In press, p. 1), similarly, portrayed the Wagner method as “rooted in the numerical taxonomy of the 1960s and 1970s”. That merely ignored yet again the same history just discussed, but Williams et al. (2010; hereinafter WEW) were more creative (WEW, p. 174)55 Of course parsimony is a criterion, not an algorithm, and is limited to computers only in the minds of authors unable to apply it themselves. : Once it [cladistics] would have been directly equated with Willi Hennig’s phylogenetic systematics... Gradually it became equated more or less directly with [Wagner] parsimony, the computer algorithm. The difference was supposed to be (WEW, p. 175; their italics): The characters on Hennig’s tree are homologies. That is, they express relationships. WEW seem to be the only authors not yet aware that Hennig’s diagrams actually show synapomorphies, which need not be homologies in the usual sense (Hennig, 1966, p. 95): In general we speak only of the homology of organs, but a ‘character’ may also be the absence of an organ... The absence of wings in the Anoplura [sucking lice] and Mallophaga [biting lice] is a synapomorphous character. In that case, the shared possession of wings in other Pterygota—certainly homology—is also symplesiomorphy relative to secondary absence, so that homologies plainly need not be synapomorphies. WEW (p. 187), however, proposed to group by homology: Relative relationships, in the sense of sameness or homology, are better ways to classify and summarize overall taxic relationships than inferring genealogies or phylogenies. As homologies can be symplesiomorphies, that was a recipe for forming paraphyletic groups. Yet not only did WEW (p. 187) pretend that their use of homologies corresponded to Hennig’s, they went on to describe their defective approach (with—I assume—unintended hilarity) as “the cladistic redux”! It was not the cladistic anything, only an absurd attempt to disguise syncretism. This explains why WEW were eager to create the deceptive impression that the Wagner method departed from Hennig’s principles. As Hennig did, Wagner parsimony can recognize secondary absences as synapomorphies (cf. Farris, 2012), and such cases expose homologies as symplesiomorphies. It also explains why WEW took the precaution of not quoting Hennig. In that connection, it should be mentioned that Nixon and Carpenter (2012) have also recently pointed out that homologies can be symplesiomorphies, their paper then being met by an indignant reaction from Williams and Ebach (In press, p. 2): Nixon and Carpenter cannot theoretically justify their claims based on their knowledge base (which, apparently is limited to two sources: Willi Hennig and James S. Farris)66 Nixon and Carpenter (2012) actually cited some 31 articles and books. . What we see here is a form of systematic malpractice — distorting history to favour a particular method. Characteristically, Williams and Ebach’s rambling discussion never actually got around to explaining just how the alleged distortions of history were supposed to favour a particular method. That remarkable level of self-righteousness seems noteworthy, however, considering that WEW apparently saw no malpractice in basing their own method entirely on a fabricated portrayal of Hennig’s approach. I suspect that for Williams and Ebach, quoting Hennig accurately will count as “distorting history”.