To the editors:

Around 520 BCE, Xenophanes observed that people’s concepts of their gods are reflections of themselves. So it is with scientists we admire. If I am suspicious of evolutionary thinking, then my scientific hero must also have been suspicious of evolutionary thinking. If I embrace evolutionary thinking, then my scientific hero must have embraced it also.

Andrew Brower and I both embrace Willi Hennig as one of our scientific heroes, but we have very different opinions on the role of evolutionary thinking and Hennig’s embrace of evolutionary thinking.¹ My reading of Hennig’s Phylogenetic Systematics reveals a scientist immersed in evolutionary thinking.² Entire parts of the book—Part III, for example—are devoted to Hennig’s thoughts on how to use phylogenetics to test what he saw were the fundamental questions about evolution in the mid-twentieth century. For example, what is the true nature of macroevolution and the origin of higher taxa? Parts of the book can be read as one long argument against idealistic morphology and the idea that characters, rather than common ancestry, define taxa. The real objective of the phylogenetic approach is not to restrict itself only to phylogenetic relationships, but to use the fruits of phylogenetic research to get at all levels of biological inquiry. The phylogenetic approach maintains that only the phylogenetic system provides a firm evolutionary foundation for comparative biology at all levels. In other words, phylogenetic systematics is good biology because it is good evolutionary biology as applied to reconstructing phylogenies and ordering diversity.

Having said that, I think I understand the suspicion of Brower and others about evolutionary thinking. The American branch of Hennigian proponents were fighting a dual battle against phenetics, which is technology-driven, and evolutionary taxonomy, which finds justification in certain evolutionary thinking. The pheneticists were fairly easy to deal with. But perhaps because evolutionary thinking was used to bolster the claim of the evolutionary taxonomists, some proponents of Hennig’s methods may have been led to assume that there was something wrong about tying the Hennigian paradigm too closely to evolutionary thinking. If paraphyletic groups, rejected as unnatural by Hennigians,³ can be justified by appeals to evolution, perhaps the problem is with evolution. So, let us distance ourselves from evolution and let cladistics stand alone, without evolutionary assumptions or at least with a minimum of such assumptions. I believe that these were the motivations of what we evolutionary Hennigians call the transformed or pattern cladists. I simply do not agree with them. To me, phylogenetics is good evolutionary biology. And if evolutionary biology needs to change in the face of phylogenetics, then change it.

Why do I object to the rejection of process, or evolutionary, assumptions? Because a process-free phylogenetics cannot be justified. Characters are not replicators. Their appearance generation after generation requires many processes, including reproduction (tokogeny) and ontogeny. Birds do not have feathers in all parts of their lifecycle. Tetrapods do not have limbs at all stages of their lifecycle. Hierarchies require tokogeny and speciation (phylogenesis), and for speciation to be a process, species need to be viewed as replicators and individuals, as Hennig viewed them. Evolutionary thinking of a kind different from that of the evolutionary taxonomists is essential to move systematics from a technology to a science. Birds do not form a group with feathered theropods because they have feathers, but because they all share a common ancestor. Feathers are simply a marker in corroboration of that relationship.⁴

Brower claims that common ancestry is a metaphysical construction with “no empirical criteria for determining its validity.” I respond that claims of common ancestry are strongly supported by the evidence. The logic of this support is lucidly described by Elliott Sober.⁵ Hennig had a different take, lucidly discussed in Olivier Rieppel’s analysis of Grundzüge einer Theorie der phylogenetischen Systematik.⁶ To Hennig, ancestors lived on through their descendants. Thus the ancestor of Aves is not only a member of Aves; it is also equivalent to all species of birds.⁷ This is because, at its origin, what is now recognized as the large clade Aves was composed of but a single species, the Ur-bird. Scientists may never discover remains of the Ur-bird, or if they do, they may not recognize it as such. But we do see the consequences of its origin. They are the feathered friends we admire today and those that lived in the past.

As a metaphysical realist, I submit that when faced with a phylogeny, one can do one of two things. Either one can conclude that common ancestors are real as an empirical result of a successful phylogenetic analysis, or one can claim that the hierarchy is a thought of one’s favorite god, stumbled upon because this favorite god happens to think in terms of clades. To reject the idea that hierarchies are very strong evidence of common ancestral species—or even parts of species, if species are merely collections of populations—is to reject the evolutionary paradigm itself.

Brower makes much of parsimony as a central tenet of cladistics. I assume he refers to the cladistic parsimony of Sober, what I term here simple parsimony.⁸ Hennig did believe that the best solution was that of the harmonious covariation of shared derived characters, or synapomorphies. And phylogeneticists have believed that the auxiliary principle—never assume homoplasy first—leads to a parsimony approach. But Hennig did not discuss the principle of parsimony directly. And he rigorously defended using a process ruled by the principle of reciprocal illumination, in order to escape circular reasoning. Brower claims that Hennig attempted to escape circular reasoning by rejecting “a priori assumptions about evolutionary processes.” Anyone who believes this claim should read Hennig’s Phylogenetic Systematics, in particular Chapter 1, more carefully. There are several kinds of parsimony, implying that there are several models of parsimony, if only informal. Sober has established that likelihood and parsimony are closely related.⁹ There is even a way to judge the parsimony of a likelihood model. The Akaike information criterion accesses the cost of adding parameters to likelihood models of any kind. It is not the phylogeny that is evaluated, but the model relative to other models. The model of simple parsimony usually employed is this: the covariation of true synapomorphies (homologies) is greater than the covariation of false synapomorphies (homoplasies), while the covariation of symplesiomorphies (shared primitive or ancestral characters) is discounted as not informative. This should result in the shortest pathway to the origin of the characters in the analysis so that tree length is minimized. I will return to why symplesiomorphies are not informative later. But that is not all. In order to sort homologies into primitive and derived characters, one needs a reference frame. Hennig described this reference frame: “Naturally these characters are ‘autapomorphous’ only if the group in question is compared with other groups.”¹⁰ This comparison is now known as outgroup comparison. The autapomorphies, or unique evolutionary innovations, of a monophyletic group are the synapomorphies shared by members of the group, including, of course, the ancestral species at the time it speciated into its first descendants. Specifically, since phylogeneticists reject any notion of ancestral groups, it is the autapomorphy of the ancestral species that lives on in its descendants as a synapomorphy or as further innovations of that synapomorphy. Using prior knowledge to solve a problem suggests a quasi-Bayesian component of the model of analysis.

That various forms of parsimony are model-based should be no surprise. All science is based on models of some sort. Simple parsimony of the type applied in many analyses has the virtue of having relatively few parameters. But when simple becomes simplistic, then the model itself no longer attends to Ockham’s razors.¹¹ Cladists’ first efforts to account for additional complexity using more complex models did not employ likelihood models but, rather, various forms of weighted parsimony.¹² Employing likelihood analysis is a way of responding to the idea that the data may be more complicated and thus require a more complex model. So, what is the justification of employing likelihood models with more parameters over simpler parsimony models with fewer parameters? The justification is that since more is known about the characters, a priori, the parameters can and should be multiplied. The a priori knowledge must be validly gathered from other fields of knowledge. A simple example follows. There are more transitions in DNA base pair changes than transversions. From current knowledge of the nature of DNA and the redundancy of the triple code, it is possible to hypothesize why transitions might be more common than transversions. It is further possible that, given enough time, more than one of the same transition mutation may occur, giving rise to homoplasy. If phylogeneticists do not take that into account, they are being simplistic and, to my mind, violating the parsimony principle. In some cases, scientists have no relevant knowledge in other fields to make additional assumptions. This is especially true of morphological characters. In such cases, simple parsimony is the right approach. A likelihood model can be employed to obtain a similar result, but that model, if examined, will be similar in parameters and assumptions to the simple parsimony model. But why likelihood? Let me express an opinion.

The fruits of science are usually expressed as a result and a statistical error associated with that result. The simple example is a mean and standard deviation. These reveal a conclusion and the uncertainty of that conclusion. In spite of efforts, the scientists who perform phylogenetic parsimony analyses have never developed a straightforward statistically interpretable error of the uncertainty.¹³ Likelihood, though, yields an uncertainty. That has a certain appeal.

Mapping characters onto a tree, as is done in classical Hennigian analysis, is transparent and is to be applauded for its transparency. The question is, are there any synapomorphies inherent on a likelihood tree or is the tree just another form of phenetics? The answer is unclear, as branch support in both likelihood and Bayesian analyses are in the form of other criteria. But consider this: any tree topology that is rooted has synapomorphies. They may be false, such as in a badly rooted tree, poor taxon selections, etc., or valid for the analysis, as in a well-rooted tree, with good taxon selection, etc. All a researcher has to do is map the apomorphies on the tree. The apomorphies can be extracted through ancestral states reconstruction with their probabilities of transition (plesiomorph to apomorph) expressed as posterior probabilities. At least one such analysis has been performed on a Bayesian tree. It yielded the same synapomorphy distribution as a tree based on simple parsimony.¹⁴ Until someone has the computer power and energy to perform ancestral states analyses on a tree of molecular data, scientists will not see molecular synapomorphies on such trees. They can, however, compare the results of those trees with the morphological data they have. Do the molecules show something different than the morphology? If so, then things get interesting.

I would like to return to the uninformative nature of symplesiomorphies mentioned above. Why discount them? After all, they are homologous with apomorphies. Sober has provided fundamental justification. There are both biological and statistical reasons as well, if one accepts the prior information provided by the outgroups. Fins and limbs are phylogenetically connected, fins being the preexisting condition that later became limbs. The condition of having fins is a synapomorphy of the larger group that includes the small group that has limbs. Pectoral fins supposedly arose once and have been counted once; pelvic fins supposedly arose once and have been counted once. In terms of descent with modification, their information content in confirming the monophyly of their respective and nested monophyletic groups has been exhausted. To use either again is as statistically unsound as counting the tallest person in the room twice when calculating the average height of people in the room.

This brings me to another, somewhat subtle issue: presence–absence thinking. Birds have feathers; lizards lack feathers. A “1” means the presence of feathers. Does the “0” mean the absence of feathers? Or does it mean a lack of information—that we do not know what covers lizards? True, one can perform a parsimony analysis on a set of characters and non-characters expressed this way. But it is a depauperate analysis. Lizards have epidermal scales, and such scales are the precursors of feathers.¹⁵ To a phylogeneticist such as myself, “0” is a code for epidermal scales. I know very well what covers lizards. Consider this statement: Tetrapods have legs (1), but fishes lack legs (0). It is accepted that legs are homologous to pectoral and pelvic appendages, as are fins. So how to express something that might make sense? As Brower states, “But the presence of paired pectoral and pelvic appendages of any kind is a synapomorphy for a larger group of vertebrates including tetrapods plus most fishes.” So, legs are pelvic appendages. This is a statement grounded in the evolutionary assumption of transformation that legs are homologous to other pelvic appendages. Transformation is involved. Now consider the parallel 0–1 statement concerning lizards (feathers absent) and birds (feathers present): But the absence of feathers of any kind is a synapomorphy of a larger group of including… what exactly? The 0–1 coding of synapomorphy and absence robs phylogenetics of evolutionary meaning and reduces Hennig’s phylogenetics to mere technology, at best. I am not sure Hennig would be amused.¹⁶