EvoWiki is now a project of the
We are moving all content to RationalWiki.
See the for details! .
Parsimony and its role in Phylogenetic Reconstruction
Parsimony in Phylogenetic Reconstruction: A Critical Look
Few concepts are as integral to science as is parsimony. The idea is largely encapsulated by Ockham's Razor, in other words, all things being equal the simplest explanation invoking the fewest ad hoc hypotheses will tend to be the correct one. It is essentially an epistemology that seeks to maximize the ability of new empirical data to differentiate bewteen hypothetically equally valid postulates (Popper 1959, Brower 2000). The "hard" sciences in particular hinge upon the validity of this axiom, and it is thus only natural that the biological sciences should do the same. The great physicist and popularizer of science, Richard Feynman, once argued that parsimony was part of the very definition of science, and I would not presume to dispute this observation.
With the advent and triumph of the cladistic method for the reconstruction of phylogenies, it has become a lamentable fact that algorithmic analysis of characters will produce a plethora of resultant cladograms, often with little demonstrable difference between them. To sort out this mess of data, systematists use the principle of parsimony to help elucidate which of the resultant phylogenies is most likely to be accurate (i.e., closest to the "true phylogeny" of the group being studied). Generally speaking, this works perfectly well. However, evolution is not so simple or so kind as to permit us the luxury of naive adherence to parsimony analysis.
As applied to phylogenetic analyses parsimony makes one explicit assumption: that homoplasy, parallelism, and character reversals, are all minimized vis-a-vis alternatives. Thus, characters are selected to have evolved the fewest number of times. The astonishing thing about this supposition, which guides so much of cladistic analysis, is that biologists have always maintained just the opposite, asserting that these factors are both persistent and permeate every aspect of organic evolution (Cain 1982, Carrol 1982, 1988, Carrol & Dong 1991). Is this an accurate observation on their part? Let us examine, as our example, the evolutionary history of Aves. The history of birds, from their inception to the Quaternary is one riddled with convergence, reversal, and parallel acquisition of characters, of which some of the more noteworthy examples are enumerated below:
b) Sphenisciformes and Plotopteridae (Olson & Hasegawa 1979, Olson 1985a, Feduccia 1996--an example of whole-body convergence).
c) Cathartidae and Gypaetinae (Sibley & Ahlquist 1990, Feduccia 1996--convergence).
f) Alcidae and Pelecanoididae (Storer 1960, Feduccia 1996--convergence).
g) Dromornithidae and Ratitae (Olson 1985a, Murray & Megirian 1998, Paine 2000--convergence, and reversal).
h) Ameghinornis minor, Aenigmavis sapea and Phorusrhacidae (Olson 1985a, Feduccia 1996--convergence).
In the defense of cladistics, it is often wrongly assumed (something that the author will freely admit to being guilty of at times), that this methodology dismisses recurrent convergence and parallelism in organismal evolution. This is not actually the case, but the concept of parsimony as applied to phylogenetic analysis does state that the phylogeny which requires less convergence and parallelism than alternate phylogenies, is more likely to be correct. Generally speaking, this is an entirely accurate assumption. But what happens when the lines between phylogenetic hypotheses are blurred, and equally viable solutions are derived from a numerical analysis, which can only be sorted using the criterion of parsimony? Shall we automatically reject those hypotheses that require comparatively higher degrees of convergence, for precisely that reason? Too often this has been the case, yet if homoplasy and its corollaries are as pervasive as all the evidence indicates, it therefore follows that we must stringently apply the concept of parsimony, using it reasonably and not dogmatically. Absent any contradicting evidence parsimony is the most reasonable and defensible argument which can be advanced, and yet seldom is this the case (White, pers. comm.). In examples where alternative phylogenies are equally viable, there is simply no justification given the degree to which convergence and parallelism exist in organismal evolution, to assume that the more parsimonious hypothesis, is the most accurate (i.e., closest to the "true phylogeny" of the group being analyzed).
We can test this directly be examining cases in which parimsony upheld a phylogenetic hypothesis which we knew was not congruent with the actual relationship of the taxa under consideration. Consider, for example, the case of "Gaviomorphae," and subsequent hypotheses which grouped Gaviiformes and Podicipediformes as a holophyletic clade.
Historically loons and grebes have been regarded as closely related, due to their superficial similarity to each other. For instance, Furbinger (1888) and Seebohm (1888b) argued that loons and grebes were sister taxa on the basis of skeletal characters. Pycraft (1899b) concurred with this assessment. Garnder (1925) disagreed with these assessments on the basis of tongue anatomy, but it was not until the work of Stolpe (1932, 1935) that a convincing argument was presented against common ancestry for loons and grebes. Stolpe demonstrated using osteological and myological characters that multiple alleged traits unifying Gaviidae and Podicipedidae were non-homologous and of superficial similarity only. He catalogued extensive differences between the two taxa which combined with fundamental lack of homology between key elements of the diving apparatus, persuaded Stolpe against a close relationship between loons and grebes (after Stolpe 1935):
a) Dorsal apterium limited to the neck in loons, and back in grebes.
b) Sternotracheal musculature symmetrical in loons, asymmetrical in grebes.
c) Paired carotids in loons, single left carotid in grebes.
d) Dorsal vertebrae free in loons, fused in grebes.
e) Caudal border of sternum notched in loons, laterall notched in grebes.
f) 11 primaries in loons, 12 in grebes.
g) Patella absent in loons, pyramidal in grebes.
h) Hypotarsus ridged and traingular and open proximally in loons, complex with with canals and grooves in grebes.
i) Webbed feet in loons, lobate in grebes.
j) Tongue possesses a patch of spinuous processes basally in loons, in grebes a single caudal row of such processes is present.
Stolpe concluded that loons and grebes were an exemplar of convergent evolution, with similar morphologies dictated by shared environmental stimuli grafted onto disparate phenotypes. Stolpe's research reached broad consensus in the ornithological community. In his 1934 analysis of avian relationships, Stresemann explicitly rejected a close relationship between Gaviidae and Podicipedidae. Hudson (1937) in a study of myology in loons and grebes verified Stolpe's observations. Mayr & Amadon (1951) in their classification of living birds erred on the side of tradition and tentatively retained Podicipedidae as the closest living relatives of Gaviidae, but this marked a notable exception from the consensus in systematic ornithology during the so-called modernist era. In his magisterial review of avian phylogeny, Wetmore (1960), though retaining Gaviidae and Podicipedidae as close relatives, rejected immediate common ancestry for loons and grebes. In a series of studies on the osteology and evolution of sea and diving birds, Storer (1960, 1971) reexamined Stolpe's data and confirmed the fundamental lack of homology between varying analogous modifications for a diving niche in both grebes and loons (e.g., the pelvic myology, the cnemial crest) and in 1971 argued for the charadriiform affinity of loons. Pioneering molecular studies using electrophoretic patterns of egg-white proteins congruently yielded data linking loons and Charadriiformes to the exclusion of Podicipedidae (Sibley 1960, Sibley & Ahlquist 1972, 1990). Observing that the bill and vertebrae of basal Sphenisciformes from the Eocene (e.g., Paleoeudyptes) bear striking resemblances to those of loons, and further commenting that detailed similarities in feather structure are present in both loons and penguins, Olson (1985a) suggested that the two may be closely related. In their contentious 1990 analysis of genetic distance data, Sibley & Ahlquist (1990) reached a similar conclusion. Thus, by the 1980s at least, a formidable amount of data arguing against recent common descent for loons and grebes had been assembled, ranging from morphological characters to the nature of their respective protein groups.
Thus came the sense of surprise and indeed deja vu when in a 1982 paper in Systematic Zoology, Joel Cracraft unveiled a cladistic analysis of Hesperornithiformes, Gaviiformes, and Podicipediformes, and concluded that these orders constituted a holophyletic clade, designated "Gaviomorphae." Cracraft rejected the hypothesis that these taxa were merely examples of convergent evolution on grounds of parsimony: it was more parsimonious to believe that they constituted a clade than believing that these three orders were convergent. The actual fact of the matter is that hesperornithids, loons, and grebes, do not share any synapomorphic characters with which to support the holophyly of "Gaviomorphae." Cracraft defended the validity of "Gaviomorphae" based on the structure of the cnemial crest, and yet in the constituent taxa of "Gaviomorphae", the cnemial crest is composed of different elements and is thus non-homologous. Stolpe knew as much in 1935, and we can thus understand the reception Cracraft's hypothesis received in Olson's (1985a) review of the fossil record and phylogeny of birds.
In Cracraft's defense, he did alter this original topology in 1986, advocating instead a gaviiform/podicipediform nexus, which has since become an enduring myth of systematic ornithology. Cracraft and other cladistic analysts have since defended what is a patently false phylogeny on several grounds, most commonly stating that mere morphological difference (ergo phenetics) is not sufficient to display convergence. This is an entirely accurate objection, except it has no bearing in this case because there exists a wealth of data to suggest that loons and grebes are more closely related to other birds than either is to each other (Sibley & Ahlquist 1990).
As elucidated above, both morphological and molecular data have consistently failed to support the holophyly of a proposed loon/grebe clade. Yet cladistic analyses continue to unite these groups based upon characters which are known to be convergent, on grounds of parsimony. Consider for instance the recent and very thorough analysis of neornithine phylogeny offered by Mayr & Clarke (2003), in which their strict consensus cladogram of the two most parsimonious trees resulting from an analysis using both ordered and unordered characters, yielded Podicipedidae + Gaviidae as a holophyletic clade. The characters advanced as synapomorphic of this clade are of great interest, as but for two they are all homplastic:
a) Sternum lacks robust spina externa rostri
b) Pelvis elongate, compressed mediolaterally
c) Femur short, and stout
d) Hypotarsus possesses medial and lateral cristae which surround the flexor tendon cannal
e) Tendon of M. flexor digitorum longus enclosed within an osseus canal
f) M. gastrocnemius possesses two heads
g) M. fibularis longus does not branch to M. flexor perforatus digiti III
h) Zygomatic process present
i) Corpus of pygostyle perforated caudoventrally
j) 17-18 vertebrae ankylozed in synsacrum
Characters a-h, are demonstrably convergent amongst diving birds due to their shared habitus, and character (h) is a reversal. We are thus left with but two characters to support the holophyly of Gaviiformes + Podicipediformes, a hypothesis so weak that Mayr & Clarke (2003) conclude that the character support for this clade is sufficiently wanting to call into question the holophyly thereof. In doing so, Mayr & Clarke are only 68 years behind the rest of ornithological science in determining that loons and grebes do not form a holophyletic grouping but in the imagination of some systematists. A similar result to earlier cladistic anaylses, even with erroneous data extracted, reveals a salient lesson. Cleaning up the characters is not sufficient in and of itself--one must focus on characters that actually are reflective of common descent and not other factors in extrapolating possible phylogenetic histories from available data.
Objections may arise to the effect that we cannot dismiss these additional traits related to the locomotor apparatus as convergent simply because one character therein (i.e., the cnemial crest), is not homologous. Yet let us examine this assertion. The cnemial crest is a fundamental component of the locomotor apparatus in any diving bird, hence its prominence in various divers (noted, indeed, by Cracraft in 1982 and used by him as a synapomorphic character). The formation and homology of the cnemial crest is related to the entire developmental architecture of the hindlimb and the lack of homology across the cnemial crests of loons, grebes, and the Hesperornithiformes, demonstrates that in these lineages the modification of hindlimb in response to a diving habitus, followed different ontogenetic pathways and constraints. Combined with the significant evidence to indicate that both loons and grebes are more closely related to other avian clades than either is to each other, defending a holophyletic Gaviidae + Podicipedidae (or Gaviidae + Podicipedidae + Hesperornithiformes) becomes untenable.
Significantly, even the removal of erroneous data (such as that used by Cracraft originally) from the data matrix (e.g., Mayr & Clarke 2003) still yields a holophyletic loon/grebe assemblage. Something is very wrong here. Either the morphology is lying to us, or parsimony is not retrieving an accurate phylogeny. The answer as to which is to blame should be quite obvious. This case stands as an exemplar of a situation in which unreasonable reliance on parsimony analysis resulted in a phylogenetic hypothesis which is suspected on other grounds to be incorrect, and is thus illustrative of the central theme of this article. The most rational conclusion is that we cannot automatically reject a phylogenetic hypothesis for the simple reason that it requires more homoplasy/parallelism than do alternatives, particularly if the differences between these contrasting hypotheses are negligible. When parsimony ceases to be a guideline and is instead elevated to an ex cathedra pronouncement, parsimony analysis ceases to be science. It is worth noting that Lee (2002) noted that such adherence to parsimony without consideration for justifying it amounts to a tautology, and argued that (2002, 218):
"A full justification of cladistics must therefore also demonstrate that parsimony is superior at representing or retrieving that hierarchy compared with other methods..."
Considering that we have demonstrable cases where parsimony has dictated an incorrect phylogeny, this objection needs to be closely scrutinized.
Brower, A. 2000. Evolution is not a necessary assumption of cladistics. Cladistics 16: 143-154.
Cain, A. J. 1982. On homologies and convergences. In: Joysey, K. A. & Friday, A. E. (eds.), Problems of Phylogenetic Reconstruction. Systematics Association Special Volume 21: 1-19.
Carroll, R. 1982. Early evolution of reptiles. Annual Review of Ecology and Systematics 13: 87-109.
Carroll, R. 1988. Vertebrate Paleontology and Evolution. W. H. Freeman & Company, New York.
Carroll, L. & Dong, Z. M. 1991. Hupehsuchus, an enigmatic aquatic reptile from the Triassic of China, and the problem of establishing relationships. Philosophical Transactions of the Royal Society of London, ser. B, 331: 131-153.
Cracraft, J. 1982. Phylogenetic relationships and monophyly of loons, grebes, and hesperornithiform birds, with comments on the early history of birds. Systematic Zoology, 31: 35-56.
Cracraft, J. 1986. The origin and early diversification of birds. Paleobiology, 12(4): 383-399.
Feduccia, A. 1996. The Origin and Evolution of Birds, First Edition. Yale University Press, New Haven.
Furbinger, M. 1888. Untersuchungen zur Morphologie und Systematik der Vogel. Two volumes. Amsterdam, Holkema.
Gardner, L. L. 1925. The adaptive modifications and the taxonomic value of the tongue in birds. Proceedings of the United States National Museum 67: 1-49.
Hudson, G. E. 1937. Studies on the muscles of the pelvic appendage in birds. Amer. Midl. Nat. 39: 102-127.
Lee, M. S. Y. 2002. Divergent evolution, hierarchy and cladistics. Zoologica Scripta 31: 217-219.
Mayr, E. & Amadon, D. 1951. A classification of recent birds. American Museum Novitates 1496: 1-42.
Mayr, G. & Clarke, J. 2003. The deep divergences of neornithine birds: a phylogenetic analysis of morphological characters. Cladistics 19: 527-553.
Murray, P. F. & Megirian, D. 1998. The skull of dromornithine birds: Anatomical evidence for their relationship to Anseriformes. Records of the South Australian Museum 31: 51-97.
Olson, S. L. 1985a. The fossil record of birds. In: Farner, D. S., King, J. R., and Parkes, K. C. (eds.), Avian Biology, Volume 8, 79-252.
Olson, S. L. & Hasegawa, Y. 1979. Fossil counterparts of giant penguins from the North Pacific. Science 206: 688-689.
Paine, S. 2000. The demon duck of doom. New Scientist 166(2240): 36-39.
Popper, K. 1959. The Logic of Scientific Discovery. Basic Books, New York.
Pycraft, W. P. 1899b. Contributions to the osteology of birds. Part IV. Pygopodes. Proceedings of the Zoological Society of London 1899(1900): 1018-1046.
Seebohm, H. 1888b. An attempt to diagnose the suborders of the great Gallino-Gralline group of birds, by the aid of osteological characters alone. Ibis 30: 415-435.
Sibley, C. G. 1960. The electrophoretic patterns of avian egg-white proteins as taxonomic characters. Ibis 102: 215-284.
Sibley, C. G. & Alhquist, J. E. 1972. A comparative study of egg-white proteins of non-passerine birds. Bulletins of the Peabody Museum of Natural History 39: 1-276.
Sibley, C. G. & Ahlquist, J. E. 1990. Phylogeny and Classification of Birds: A Study in Molecular Evolution. Yale University Press, New Haven.
Stolpe, M. 1932. Physiologisch-anatomische Untersuchungen uber die hintere Extremitat. Journal fur Ornithologie 80: 161-247.
Stolpe, M. 1935. Colymbus, Hesperornis, Podiceps, ein Verglich ihrer hinteren Extremitat. Journal fur Ornithologie, 80: 161-247.
Storer, R. 1960. Evolution in the diving birds. Proceedings of the Twelfth International Ornithological Congress, 694-707.
Storer, R. 1971. Adaptive radiation of birds. In: D. S. Farner, and J. R. King (eds.), Avian Biology, 149-188
Stresemann, E. 1934. Aves. In: Kukenthal, W. & Krumbach, T., (eds.), Hanbuch der Zoologie vol. VII, pt. 2.
Wagner, G. P. & Gauthier, J. A. 1999. 1,2,3=2,3,4: a solution to the problem of the homology of the digits of the avian hand. Proceedings of the National Academy of Sciences 96: 5111-5116.
Wetmore, A. 1960. A classification for birds of the world. Smithsonian Miscellaneous Collections 139(11): 1-37.
Compiled by JGK, who would also like to thank an anonymous reviewer for commentary and suggestions, as well as the kind editing of Richard White.