Multivariate Approaches to Development and Evolution

Development is the usual arena of evolutionary change in morphology, and much of evolution obviously depends on it. It is very common, however, for discussions of human evolution and developmental change to focus on individual characters (brain size, body weight, bipedalism, etc.). This particularist approach has been enormously useful in understanding how individual phenotypic modules have changed in the course of human evolutionary history. But it has also led to dilemmas when from the pattern of developmental evolution of individual features broad statements about evolution in general are made (e.g., paedomorphosis vs. peramorphosis). Dissociated heterochrony (McKinney and McNamara 1991) and heterotopy (Zelditch and Fink 1996) are natural consequences of the modular nature of most organisms, and they have often been documented. The relative dissociability of phenotypes (quasi-independence of Lewontin 1978) justifies a more comprehensive treatment of developmental change whenever synthetic statements are sought. Conversely, high morphological integration may well entail uniform developmental change across characters. A multivariate perspective permits the discernement of such alternatives, and the production of average descriptions of development and evolution in multidimensional space. Two approaches are immediately obvious by contrast to the particularist approach: the tabulation of how developmental change occurs across individual characters (a useful line of attack which often figures in the literature, but which soon becomes intractable as the number of characters increase and as redundancy confounds the potential for synthetic inferences), and the direct summarization of multivariate variation with statistical techniques. This paper will address the latter. McKinney and McNamara (1991) and Tissot (1988) give good overviews of the advantages of multivariate analysis in the context of heterochrony, and Shea (1985) and Klingenberg (1996) discuss multivariate allometry. While I will elaborate on traditional multivariate methods in this paper, my intent is to eventually move beyond approaches which have now become de rigueur, such as principal components analysis, and explore alternative ways of assessing patterns and processes of human developmental change in a multivariate framework. Evolutionary anthropology as a discipline, both in biological and paleontological terms, shares much with evolutionary biology and paleobiology, and yet for a plethora of reasons (both sociological and historical) has remained self-contained and somewhat isolated from methodological and conceptual developments in those fields. The reverse is also true, but from the standpoint of anthropology this isolation has meant a rather idiosyncratic application of such developments. This is because they are only relevant to the extent that they can be fruitfully adapted to the empirical situations evolutionary anthropology is usually faced with, and tied to the tradition of empirical research in a straightforward fashion. Landmark-based approaches, for example, are finally gaining wide acceptance within anthropology (e.g., Lynch et al. 1996; Chaline et al. 1998). Still, a number of recent advances in multivariate biology and paleobiology have not yet been explicitly applied in evolutionary anthropology. Most discussions of evolutionary change in development focus on the fundamental notion of ontogenetic trajectory -- how it is modified from ancestor to descendant. If one had access to a fully resolved phylogenetic tree, and if complete ontogenetic sequences were available for all species, one would proceed to interpret transition after transition in terms of how particular trajectories were modified. This tradition, while defensible (it will be applied below), hides the possibility that other observables might be informative in understanding the relationship between development and evolution. For example, developmental constraints are usually taken as a given and their potential breaking deemphasized, since heterochrony is usually thought to occur along an established trajectory. A focus on individual transitions also impedes a comprehensive appraisal of sweeping cross-phylogenetic predictions, such as Von Baer's laws. I will argue below that an ensemble approach, explicitly directed at identifying statistical regularities across a clade, may give insights that would not be immediately obvious to a student of individual transitions. From the perspective of human evolution studies, this implies incorporation of principles and methods most commonly found in invertebrate evolutionary paleobiology. In particular, the concepts of morphospace and disparity (Raup 1966; Foote 1997) are implicit in much of the data and aims of human evolution studies -- their explicit recognition could add new dimensions of intepretability. By adapting the notion of morphospace and disparity to the arena of development (see Eble 1998a, 1999a), I intend to suggest how such concepts can be valuable in the study of human developmental change and, by extension, human evolution in general. Heterochrony is a powerful mechanism of evolutionary change in development. However, as much as in a univariate context (Rice 1997), in a multivariate context the question of whether additional ways of studying developmental evolution are possible or useful deserves more attention. Ultimately, all changes in development must involve changes in rate or timing (whether uniform or non-uniform, along a single trajectory or leading to novelty, having a spatial component or not) -- the question is whether any given formalism should be used by default or whether a more pluralistic attitude is desirable. Current discontent with the standard nomenclature of heterochrony has been for the most part cast in terms of univariate situations, in line with the received typology (Alberch et al. 1979). Exceptions include Zelditch and Fink (1996) and Godfrey and Sutherland (1996). While multivariate heterochrony will inevitably surface in the discussions, this paper will mostly be concerned with the general utility of multivariate approaches to developmental change, on the one hand, and with the possibility of, through them, studying evolution and development in novel ways, on the other. I will proceed by first discussing the general concepts of morphospace and disparity. I will then briefly review some techniques for the multivariate description of variation. The concepts of developmental morphospaces and developmental disparity will then be introduced. Finally, I will apply such concepts and some of the techniques in an empirical illustration relevant to human evolution: a reanalysis of Heinz's (1966) dataset on growth in extant hominoids. To appear in Human Evolution through Developmental Change, edited by K. McNamara and N. Minugh-Purvis. Baltimore, MD: Johns Hopkins University Press.