Shape, Morphology and Phylogeny Symposium: A Review
A symposium held at the Second Biennial meeting of the Systematics Association
University of Glasgow, August 25, 1999.
Morphomeritcs is a discipline that occupies the conceptual vertex between geometry, numerical analysis, and systematics. While morphometrics has traditionally-and perhaps rightly-drawn most its vigour from the mathematical sides of this territory, the cost of this emphasis as been an increasing estrangement from what is perceived as the research frontiers of systematics, ecology, development, etc. In this sense many systematists view morphometrics as a set of tools in search of a problem. However, over the last few years several developments have reopened the question of the utility of morphometric data within the systematics community. These include the development of numerical methods designed to remove phylogenetic covariation from quantitative morphological data (e.g., Felsenstein 1973, 1985, 1988; Harvey and Pagel 1991; Gittleman and Luh 1992), phylogenetic methods designed to infer ancestral morphologies from quantitative data (e.g., Harvey and Pagel 1991), the use of morphometric results in phylogenetic analyses (e.g., Zelditch et al. 1995; Rohlf 1998), and the issue of the use of continuous variables in phylogenetic inference (e.g., Theile 1993; Rae 1998). The "Shape, Morphology and Phylogeny" Symposium was convened at the Biennial meeting of the Systematics Association on 25 August 1999 in order to gather the principal participants in these discussions together with other systematists who have a long-standing interest in these issues for the purpose of discussion and debate over (1) the role of quantitative morphological data in phylogenetic analysis and (2) the role of phylogenetic data in morphometric analysis.
The symposium's program occupied the entire day and was roughly divided into a consideration of role of quantitative morphological data in phylogenetic analysis in the morning session and the role of phylogenetic data in morphological/morphometric analysis in the afternoon (though there were many instances of authors addressing both topics throughout the day). After introductory remarks by Peter Forey (co-convenor along with myself) the symposium began with keynote presentations, one by Fred Bookstein and the other by Joe Felsenstein.
Bookstein's presentation ("Creases in Deformations: the Missing Link Between Morphometrics and Phylogenetics") began the symposium with a discussion of the ways homology-in the form of a discontinuity between alternative morphological states-might be recognized in morphometric data. After developing the algebra of discontinuous shape transformations, first in the one-dimensional case, then in the two-dimensional case, and finally in the context of thin plate splines (tps's), Bookstein showed how the variation of parameters that express the "directed amplitude" of the shape transformation can be used to identify local maxima (that he termed "creases" and which are akin to Rene Thom's "catastrophe surfaces") in regions of a tps where the form comparison is undergoing relatively large shape transformations. After discussing examples of creases and cautiously venturing the provisional interpretation that these features are the expression of the discontinuous morphological character states beloved of phylogenetic inference, Bookstein concluded with an offer to work with systematists to explore the crease phenomenon in general systematic and phylogenetic contexts.
Felsenstein's keynote presentation ("Statistical Inference of Phylogenies from Morphology, Including Morphometrics") complemented Bookstein's and greatly expanded the scope of the discussion by reiterating many of the points made in his highly-influential American Naturalist (1985) and Annual Review of Ecology and Systematics (1988) papers. Those papers, which pointed out that phylogeny must be taken into account whenever qualitative or quantitative comparisons between organisms are being made (and especially if the results of those comparisons are to be statistically compared to null or random models), also contained several insights into the nature of the data that can be used in phylogenetic inference that have been overlooked. Specifically, Felsenstein argued that the restriction of phylogenetic inference methods to the use of discontinuous data (e.g., characters and discontinuous character states) is a historical by-product of the development of parsimony-based algorithms. [Note: in the discussion following Felsenstein's presentation Jim Rohlf pointed out that one important aspect of this history was the desire to make parsimony-search algorithms as fast as possible and that use of discontinuous variables (which could be represented by integers) represented one way to increase tree search speeds.] In Felsenstein's view the use of continuous variables (including morphometric variables, so long as they meet minimal criteria as putative homologues) is possible under both parsimony and maximum-likelihood phylogenetic inference models.
My own presentation ("Phylogenetic Signals in Morphometric Data") followed Felsentsein's and was unfortunately marred by the room technician's failure to get the computer projection system working. I (and the audience) persevered and, using a set of backup graphics, I attempted to employ a series of practical examples to show how important the relation between phylogenetics and morphometrics is. An example analysis drawn from the MacLeod and Rose (1993) study revealed how much phylogenetic covariation there can be in morphometric data as well as how much discrimination between shape groups can be improved by eliminating the phylogenetic signal from such data. In addition, a series of morphometric-phylogenetic analyses on an example trilobite dataset showed that while morphometric analyses of gross character complexes (e.g., cranidial shape) can fail to recover a consistent phylogenetic signal, the morphometric analysis of individual morphological features-which in most cases form the basis of phylogenetic inference-can be crucial in assigning taxa correct character states as well as in recognising and describing new characters.
After tea the symposium continued with an expanded consideration of the use of morphometric data in phylogenetic analysis. Donald Swiderski's presentation ("Comparability and Homology of Morphometric Data") recapitulated the Zelditch et al. (1985) approach to identifying characters and character states using principal warp scores and emphasized (1) the crucial role of landmark points in ensuring that corresponding-and therefore putatively homologous-aspects of the form are being compared, and (2) that morphometric comparisons in systematics should be polarized in the same way that phylogenetic characters are: reference forms should be the "outgroup." While the role of the reference shape in providing a common basis for the comparison of shapes distinguishes the methods Swiderski et al. find useful in phylogenetic inference from the more traditional multivariate methods (in which the reference shape is implicitly determined and subject to change with the inclusion of any new data) the crucial issue of how reference shapes are to be maintained and what happens if they become modified in the light of new data remains to be addressed.
Jim Rohlf's presentation ("Geometric Morphometric Methods in Systematics") focused on an explanation of the non-linearity of shape space and the roles of the tangent plane and reference shape in providing means by which to represent morphological ordinations. On the topic of the relation of morphometrics to phylogenetic inference Rohlf showed how it is possible to separate the 'reference shape' used to define the tangent space for the various computations from a 'starting form' used to visualize the estimated shapes of the hypothetical (e.g., ancestral) taxonomic units (htu's). This is a very important conceptual step in that it adds a great deal of flexibility to the manner in which morphometricians can realize shape comparisons. Rohlf also demonstrated how estimated shapes could be visualized as a deformation of an average image and mentioned how images could also be estimated for the ancestral htu's using recent modifications to his TPStree program.
Tim Cole followed on from Rohlf's presentation by describing a bootstrap-based method for assessing the degree to which phylogenetic covariation is present in a set of morphometric data ("A Bootstrap Approach to Detecting Phylogenetic Signals in Morphometric Data"). As pointed out by several speakers in the morning's session phylogenetic covariation can represent a major (and largely ignored) "nuisance factor" in the interpretation of morphometric results. Cole and his co-authors offered a null model and generalized statistical test for identifying the presence of this factor in morphometric data that can be applied to the results of any morphometric study. Coles's method was illustrated in the context of this presentation using results from an Euclidean Distance Matrix Analysis (EDMA) of primates with the phylogenetic tree coming from molecular data.
David Polly closed the morning session with an investigation into the phylogenetic/morphometric implications of "Eldredge's Engima" ("Morphometrics, Comparative Analysis, and Fossils: Eldredge's Enigma Explored"). This enigma has to do with the ambiguity inherent in the cladistic convention of treating all otu's as terminal taxa on a cladogram. Under this convention the nodes of a cladistic hierarchy represent hypothetical ancestors whose character states are inferred from the state distributions of the terminal taxa. The problem comes when trying to statistically compare between-otu shape where it is necessary to know (1) phylogenetic branching order, (2) the expected per-unit-time change in shape under a random model, and (3) the actual amount of temporal divergence separating the species being compared. Factor 1 can be provided from a traditional cladistic analysis. Factor 2 can be estimated from the data (presuming there is a sufficiently phylogenetically broad set of shapes). Factor 3, however is made difficult because of Eldredge's Enigma (both in placing terminal taxa when they are fossils and in figuring divergence times). As Polly pointed out, this problem has many similarities to the problem of estimating ancestral character states.
After lunch (during which the photo that adorns this review was taken) the afternoon session commenced with Chris Humphries' presentation on homology and continuous variables ("Homology, Characters, and Continuous Variables"). Humphries agreed (in principal) with previous symposium authors that the problem with using morphometric data in phylogenetic analysis has little to do with morphometrics per se, but is created by the difficulties inherent in treating continuous variables as (putatively) homologous characters and character states. This difficulty was traced to a historical ambiguity in the concept of a "character" and the relation of "characters" to "homologues" (e.g., the discrepancy between taxic and transformational homologies). Since morphometrics has no tradition of developing (and geometry no need to develop) analytic approaches that take explicit recognition of ancestor-descendant hierarchical organizational patterns into consideration, Humphries was pessimistic about using morphometric concepts of positional correspondence as raw input into a phylogenetic analysis. Nevertheless, he did not preclude the development of morphometric methods that would incorporate such concepts.
Todd Rae took a somewhat different view of the utility of morphometric data in phylogenetic contexts. His presentation ("Scaling, Polymorphism, and Cladistic Analysis") assumed that the controversy between continuous & discontinuous characters had been resolved by systematists' routine use of continuous morphological variables (albeit after coding into implicitly discontinuous character states) in their analyses. Rather, Rae took up the point that the univariate morphometric characters used by most practising systematists (e.g., linear distances) are not independent because they are strongly correlated with generalized body size. Rae argued that, despite long-standing and well-known problems the use of ratios represented a practical and preferable alternative to use of the raw data and that this approach, coupled with appropriate coding methods, could be used to resolved several inconsistencies in contemporary systematic practice.
Karen Sidwell and Geraldine Reid focused on these so-called "gap coding methods" and provided an interesting botanical perspective in their presentation ("Testing Continuous Characters in Cladistic Analysis"). Through direct comparison of the different methods (simple gap coding, generalised gap coding, segment coding) on the same dataset they were able to demonstrate that coding methodology matters in complex and often counter-intuitive ways. In particular, they emphasized the proliferation of molecular plant phylogenies in recent years (which can be used to independent assess the patterns of morphological evolution) and the plasticity of the plant phenotype (which virtually requires the use of continuous variables when describing morphological attributes). Overall, these authors saw no alternative to the study of morphological evolution in plants than by the use of morphometric approaches.
Andrea Webster and Andy Purvis returned to the topic of inferring ancestral character states in their presentation ("Methods of Estimating Ancestral Characters") where they provided a much-needed perspective on differences between three evolutionary models (linear parsimony, squared change parsimony and Brownian motion with a single rate estimated by maximum likelihood) to constrain estimates of morphological evolution from morphometric data. Results indicated that simple linear parsimony performed better with respect to the primate dataset used by these authors.
Finally, Mark Pagel delivered a provocative summary of the afternoon's proceedings in his presentation ("Morphometrics and Phylogenetics") by pointing to a future in which phylogenetic tress are regarded as prerequisites to the understanding of morphological evolution through morphometric data. Pagel presented a series of analyses in which a generalized least-squares approach was used to model the morphological history of various clades and modern organisms used to evaluate the results of such models. His closing slide was an image of a dinosaur and his closing argument was that, by using the data encoded in both phylogenetic and morphometric approaches systematists are developing the tools they will need to better "see" the morphological history of life; a fitting end to an insightful and rewarding day.
Following the final presentation the speakers and entire audience engaged in a wide-ranging discussion that helped to link the various themes that had been presented throughout the day. In this discussion the keynote speakers set the correct tone with Fred Bookstein providing an appropriate historical perspective (rightly pointing out that while morphometrics and systematics had seemed to be growing apart in recent years, much of what was heard at this symposium bespoke an envigourated interest in the use of morphometric data by systematists) and Joe Felsenstein proposing a "current status" poll (by asking each speaker to place himself/herself on a spectrum of unrestricted-restricted-no use of continuous-morphometric variables in phylogenetic inference). From there the discussion moved off in several directions to cover philosophical stances mediating approaches to phylogenetic inference, the degree to which morphometricians should become involved in questions of phylogenetic inference (as opposed to keeping a focus on using phylogenies generated by whatever means to interpret their data, the various practical difficulties presented by different types of organisms (e.g., plants vs. animals), etc. Naturally no final consensus emerged. However, I think it is safe to say that everyone who participated in and/or viewed the program came away with a sense that the relation between morphometrics and phylogenetics contains much fertile ground for future research and, in a very real sense, is capable of addressing many of the biology's "big issues" in novel and compelling ways.
A proceedings volume is planned and will be published as part of the Systematics Association's continuing series. Look for it in late 2000-early 2001. Updates on the volume's progress along with a variety of resources from the symposium (e.g., my much talked about but never actually seen PowerPoint presentation) will be available on the symposium web site.
Felsenstein, J. 1973. Maximum Likelihood estimation of evolutionary trees from continuous characters. American Journal of Genetics 25: 471-492.
Felsenstein, J. 1985. Phylogenies and the comparative method. American Naturalist 125: 1-15.
Felsenstein, J. 1988. Phylogenies and quantitative characters. Annual Review of Ecology and Systematics 19: 445-471.
Gittleman, JL, Luh, H-K. 1992. On comparing comparative methods. Annual Review of Ecology and Systematics 23: 383-404.
Harvey, PH, Pagel, MD. 1991. The Comparative Method in Evolutionary Biology. Oxford; Oxford University Press.
MacLeod, N, Rose, KD. 1993. Inferring locomotor behavior in Paleogene mammals via eigenshape analysis. American Journal of Science 293-A: 300-355.
Rae, T. 1998. The logical basis for the use of continuous characters in phylogenetic systematics. Cladistics 14: 221-228.
Rohlf, FJ. 1998. On applications of geometric morphometrics to studies of ontogeny and phylogeny. Systematic Biology 47: 147-158.
Thiele, K. 1993. The holy grail of the perfect character: the cladistic treatment of morphometric data. Cladistics 9: 275-304.
Zelditch, ML, Fink, WL, Swiderski, DL. 1995. Morphometrics, homology, and phylogenetics: quantified characters as synapomorphies. Systematic Biology 44: 179-189.
Dr. Norman MacLeod, Micropalaeontological Research,
Department of Palaeontology, The Natural History Museum, Cromwell Road, London, SW7 5BD