J. theor. Biol. (1990) 147, 43-57

An Exercise in Rational Taxonomyt

M A E - W A N HO

Developmental Dynamics Research Group, Open University, Walton Hall, Milton Kevnes MK 7 6AA, U.K.

(Received on 24 July 1989, Accepted in revised form on 14 May 1990)

The quest for a rational taxonomy of living forms began in the 17th century. Since the general acceptance of Darwin's theory of descent with modification, however, students of morphology became preoccupied with a systematics based on the genealogy of groups; and the rise of molecular phylogenies in recent years results in a further decline in the science of morphology. Reconstructing phylogenies by itself brings us no closer to the goal of rational taxonomy, which is to uncover the natural order inherent in the forms of living things. It is proposed that the rational taxonomy of forms should be derived from a study of development, much as von Baer had envisaged. To illustrate the method, a set of segmentation abnormalities in Drosophila larvae (previously exposed to either vapour) is considered, which can be individually classified as distinct disturbances in the process responsible for establishing normal segmental pattern. The process consists of a hierarchy of four successive bifurcations dividing the embryo's body first into two parts, then four, eight, and finally 16 subdivisions or segments. This gives rise to a taxonomic map of all possible transformations which contains the "~phylogeny'" of the actual forms and provides a natural system for classifying them. Attempts to recover the "true" phylogeny by various numerical methods are summarized and their implications for the validity of the basic assumptions of contemporary systematics discussed.

The search for the order of things began in earnest the West dur ing the 17th century

(Foucaul t , 1970). A major concern was to discover a taxonomy, or a system for classifying living things that cor responded to the natural order. Such a system, according to Linnaeus , was one in which "all the difference in the character would cor respond to difference of the same value in the p lant ' s general s tructure; and in which, inversely, all the ind iv idua ls or all the species grouped together unde r a c o m m o n character would in fact have the same relat ion of resemblance in all and each of their parts".

However, op in ions differed as to how to arrive at such a system of classification.

There were two main schools of thought as regards the proper procedure to adopt . The first was the system approach, associated with L innaeus himself, and the second, the method approach, associated with Buffon, among others. The method approach consisted in making total compar i sons of every character between groups so that step by step the es tab l i shment of all identi t ies and differences is accompl i shed and

+ Paper presented at Conference on Fundamental Problems of Evolution, U.S.S.R. Academy of Science, Moscow, April 20-26, 1989.

43

0022-5193/90/021043 + 15 $03.00/0 © 1990 Academic Press Limited

44 M.-w. HO

the natural system will automatically emerge. As Adanson urged, it was necessary to search in nature herself for her system, if it is true that she really has one (see Nelson, 1979).

The system approach, on the other hand, consisted in the selection of a finite and relatively limited number of characters whose variations and constancies may be studied in any individual entity that presents itself. This system was artificial, but Linnaeus was convinced it was absolutely necessary to establish an artificial system before one could possibly find the way to the natural one.

This contrast in approaches roughly parallels that between present-day cladists, deriving from Hennig (1966) and their antagonists, the evolutionary taxonomists (e.g. Mayr, 1963). The pheneticists, the third major school of contemporary system- atics (e.g. Sokal & Crovello, 1970), are closely aligned with the cladists in basic philosophy, although not in methodology. Patterson (1988) regards it to be the opposition between pattern and process. Evolutionary taxonomists have traditionally employed an intuitive approach to selecting the most important characters for classification, often based upon hypothetical processes or "'adaptive pathways" of evolutionary transformation that have little or no independent empirical basis (Nelson & Platnick, 1984). Cladists, on the other hand, compare as many characters as possible without such assumptions. The claim is that the eventual pattern speaks for itself, and the real process will become transparent; the real process, in this case, being descent with modification (Patterson, 1988).

Thtis, a considerable gulf has opened up between all present-day systematists and their pre-Darwinian counterparts, whom I shall refer to as rational taxonomists. Whereas the original aim was to discover the natural order of living things, the preoccupation today is to search for the true phylogeny, or the genealogy of groups (see Hull, 1970). If genetic relatedness is the object of the exercise, comparisons of DNA would yield the most direct answer (Ho, 1988). After all, genealogy is defined on genetic relatedness, in other words, on the similarity of D N A sequences; irrespec- tive of whether the neutral mutation theory is true. Patterson (1988) also argues that "sampling of genotypes results in an unambiguous concept of relationship", but justifies that on the basis of the neutral mutation theory. But, as Darwin himself (cited in Hull, 1970) pointed out, genealogy by itself does not give classification. Many different classifications may be derived from a given genealogy; and con- versely, an indefinite number of different genealogies are compatible with any one classification (Hull, 1970), depending on the criteria used in translating between the two. While pheneticists and cladists both claim that their respective methods give non-arbitrary results in classification, these claims are disputed by their opponents (Mayr, 1981). The question we must address is whether reconstructing phylogenies necessarily brings us closer to the goal of the original project: that of uncovering the natural order inherent in the forms of living things.

Biological forms are generated in development, an orderly and hierarchical process which in turn imposes an hierarchical structure on the adult organization of living things. Such considerations led von Baer (1828) to suggest that the development of organisms is part and parcel of the natural order of things, and hence a natural system of classification should be devised on the basis of the most general laws of

A N E X E R C I S E ] N R A T | O N A L T A X O N O M Y 45

development. More explicitly, I propose to define the natural system of classification of forms as one based on the processes generating those forms in a forthcoming paper.

Does reconstructing phylogenies necessarily recover ontogentic order? Patterson (1983) answered in the affirmative, although the modern cladogram of gnathostome vertebrates he used as an example of phylogeny showed considerable differences from von Baer's scheme based on ontogeny (see yon Baer, 1828: 25, fig. 9). The cladogram also contradicted von Baer's first law: that the more general characters appear earlier in the embryo than the more special: thus, the amnion and allantois, characterizing Amniota, a subgroup of Tetrapoda, appear before any feature of the latter group.

Surely the question of whether phylogeny recapitulates ontogeny in the yon Baerian sense cannot be decided by a mere assertion of congruence between phylogeny and ontogeny. History has no obligation to be orderly, and developmental order, if it exists, may not survive the contingencies of history. The problem is exacerbated by the tendency of neo-Darwinists to forego form and development in favour of genes and random variation. For that implies there is no natural order, and hence classifications can only be arbitrary.

In my opinion, the project of rational taxonomy is to discover why it is that morphologies arrange themselves as orderly hierarchies of groups. Our primary conjecture is that the processes generating form and transformations are rational or orderly, and furthermore, they can be revealed by a study of development. Thus, we agree with Nelson's (1978) conclusion that '+systematics and comparat ive anatomy are possible only to the extent that ontogeny is orderly and that the concept of evolution is an extrapolation, or interpretation, of the orderliness of ontogeny". But we treat it as the starting point of our project. Tl'ie fundamental importance of development for evolution does not consist in the use of ontogenetic transformations for reconstructing phylogeny, it is that evolutionary transformations themselves can ultimately be understood in terms of developmental transformations that can be empirically investigated and that this in turn provides us with the criteria for a natural system of classification. Of course, it is possible to insist that phylogenetic relationship is the only valid criterion for classification, but then that classification would not correspond to the natural system originally conceived by pre-Darwinian morphologists , as I hope to demonstrate, in a forthcoming paper.

These considerations form the basis of a strategy for rational taxonomy (see Ho, 1986, 1988). Phylogenetic reconstructions are best achieved by comparing DNA sequences; but the t axonomy of forms can only be derived from developmental studies coupled with mathematical modelling. The purpose of the latter is to establish empirically a map of neighbouring forms such that the " t ransformational distance" between any two forms can be read at a glance. This is the space of possible forms against which evolutionary hypothesis could be tested. Our main conjecture is that evolution is governed, not by accidents, but by the dynamics of well-ordered processes, and it is that which preserves the hierarchy of development that ultimately orders the phylogeny of forms. One prediction is that evolutionary change will tend to occur between nearest neighbouring forms on the taxonomic map. [This has been referred to as the principle of minimum increase in complexity by Saunders & Ho

4 6 M . - W . H O

(1981).] Another expectation is that the frequency of a particular morphological convergence should be directly proportional to its dynamic accessibility, which would explain both the ubiquity of parallel evolution in all lineages and the widespread occurrence of particular convergences. Finally, the taxonomic map of possible ontogentic transformations is the closest approach to the realization of a natural system of classification of organismic forms that both Linneaus and yon Baer had in mind.

To illustrate the strategy, let us consider a set of developmental data extensively described and characterized by Ho et al. (1987). It consists of numerous segmentation abnormalities obtained by exposing Drosophila embryos to ether vapour at various ages between 1-3 hr after fertilization, which were then allowed to develop until the time they would normally hatch as first instar larvae. A wide variety of new morphologies were generated, presenting to us a problem in classification of the sort that systematists experience when they come across new organisms in nature.

At first, it seemed obvious that one should list all characters exhaustively, then carry out a detailed statistical analysis in order to find the best classification. However, it soon became clear that it was often impossible to establish what the characters were. Many of the disturbances involve global reorganizations of the segmental pattern in such a way that detailed homologies with normal structures could not be made. In most cases, the abnormal segments bear no resemblance to any normal segment, and their identity can only be inferred by unaltered segments flanking them. As many of the abnormal phenotypes are similar to those described as morphogenet ic mutants, we decided on a scheme of classification used by the geneticists. The resulting groups were far from natural, and were moreover, extremely heterogeneous (see Ho et al., 1987).

I now show that practically all observed morphologies can be individually classified as disturbances occurring at different stages during the developmental process responsible for establishing the normal segmental pattern. This process is suggested by our own data, and is also widely held by Drosophila geneticists studying the morphogenet ic mutants (see Akam, 1987 and references therein). The latter are classified into a hierarchy of genetic categories whose spatial effects follow a pattern of decreasing wavelengths in the embryo 's body. They can be described as a series of frequency doublings which represents the development of an ascending harmonic sequence of standing waves. Kauffman et al. (1978) were among the first to suggest this model for Drosophila segmentation. Recently, Goodwin & Kauffman (1989) analysed the available gene transcript data, and proposed a detailed model which explains the distinctive periodicities of the different gene transcripts as well as their morphogenet ic effects. A three-dimensional simulation of their hierarchical dynamic model confirms the stability of the bifurcation patterns. The model used here is the most general of the genre.

There are two generative rules. The first is a hierarchy of four successive bifurca- tions dividing the embryo 's body first into two equal parts, then four, eight, and finally 16 subdivision or segments. The second rule is that each bifurcation can only take place if the previous bifurcation is complete. The action of ether consists in suppressing the formation of bifurcations that happen to be occurring at the time

A N E X E R C I S E I N R A T I O N A L T A X O N O M Y 47

of exposure (see Ho et al., 1987). This then allows us to construct a taxonomic map of all possible ontogentic transformations. In fact, we can give each form a code name which conveys immediately its distinctive structure (see Fig. 1).

Let the letters, A, B, C, D denote the bifucations that occur successively in time, as well as the forms that have attained the particular bifurcations. Numbers fol lowing each letter identify the specific boundaries that have failed to form in the particular bifurcation. For example, B1 means that bifurcation A was complete, but B failed

A

_ _ J , d

B

1

C CI C2

@ @ @ . . CI2 C25

Cl25 C254

f O DI D2

DI2 D25

DI23 D234

BI

d2> l

BIC5

BIC4

BIC

_ _ t BICD

BICD5

BICD56

B2

l _ _ f

B2CI

FIG. 1. A taxonomic map of possible ontogentic transformations (see text).

48 M.-W. HO

in the first boundary, counting from the anterior; similarly, C23 means that bifurca- tions A and B were complete, but during bifurcation C, the second and third boundaries failed. B1CD5 means that bifurcation B failed at the first boundary, C followed as normally as possible, that is without C1 and C2 (which need not be entered); and then D (without D1-4, because of previous bifurcation failures, which again need not be entered), with the additional failure of the fifth boundary. In this way, one can enumerate all possible transformation, which come to a total of 676. This is much smaller than the number of possible forms if each segment were free to vary independently, which is 216, o r more than 60 000.

The taxonomic map is the phenotypic space within which the phylogeny of the actual forms traverses. We expect to locate all existing forms within this space; and we can also predict the existence of novel forms not yet observed.

The alignment of the bifurcation boundaries with the segmental pattern is presen- ted in Fig. 2: three head segments, three thoracic and ten abdominal segments. This alignment is the same as that used by the geneticists for the banding patterns of the gene transcripts (Akam, 1987). [The relationship between the genetic and ether- induced phenotypes is considered in detail in a separate paper (Ho & Goodwin, in preparation).]

I / / / / / / / / / J l TI T2 T5 AI A2 A3 A4 A5 A6 A7 A8 A9 AIQ

FIG. 2. The alignment of bifurcation boundaries with the segmental pattern of the first instar larva at hatching.

A sample of the data consisting of at least 30 forms were analysed and identified according to the scheme. None of these showed morphologies which are explicitly forbidden by the developmental process; in other words, all were located within the taxonomic map. For example, " ' forbidden" forms include all those with T2 and T3 without T1, those in which head structures are complete without T1, the presence of A8 without the filzkorper, and so on.

Twenty-four morphologies (Fig. 3) were selected as the forms for which a "ontogenet ic tree" is drawn (Fig. 4). The actual or observed forms are included

> Z

/-n

X

i-rl

> ..q

Z

> t--'

> X

0 Z 0 .<

FIG

. 3.

T

he

ob

serv

ed

mo

rph

olo

gie

s,

the

inte

rpre

tati

on

o

f th

e tr

ansf

orm

atio

ns

and

th

eir

ind

ivid

ual

co

de

nam

es.

4~

50 M . - W . HO

o

Y d

L/.

A N E X E R C I S E I N R A T I O N A L T A X O N O M Y 51

ta.

"l O

FIG

. 3

--co

nti

nu

ed

A N E X E R C I S E I N R A T I O N A L T A X O N O M Y 53

FIG. 4. The ontogenetic tree of the observed morphologies with hypothetical intermediates (dotted outline) based on the developmental rules.

with the hypothetical intermediates. This tree traverses the taxonomic map, but is not identical to it, having incorporated all the contingencies of history. Yet, the hierarchical structure is preserved because of our knowledge of the developmental process that is empirically and independently acquired. We may note that it is a von Baerian tree in the sense that the more general characters (domains containing multiple neighbouring segments) appear earlier in the embryo than the more special (domains corresponding to single segments); and the less general characters are developed from the more general.

A natural system of classification from this tree gives a single "order" with two "families": two " 'genera" in the first and eight in the other. We can predict the existence of a third " fami ly" and many potential "species" to be discovered could be assigned unambiguously. This system of classification is, so to speak, stable against the discovery of new forms.

It might be objected that a study of developmental teratology can have little implications for systematics, which concerns the classification of genetically distinct organisms occurring in nature. In answer to that, we point out that organisms evolve through changes in development that give rise to new morphologies. The set o f morphologies available is determined by the developmental system. Indeed, it is

54 M.-W. HO

now generally accepted (as well as a commonplace observation) that similar mor- phologies can result from genetic mutations or environmental perturbations (see Goldschmidt, 1940, Ho, 1984; Ho et al., 1987). Thus, whether genetic changes are involved is irrelevant to the problem of classification of the morphologies themselves.

At the time of exposure to ether fumes, the embryos were at different bifurcations in the patterning process. This in turn gave rise to different new morphologies. Not all possible morphologies survived to the stage at which they were compared, due to historical contingencies involving the particular batch of embryos, the specific manner of ether exposure, and no doubt, the different survival rates of individual morphological forms. This is the stuff of phylogenetic transformations, if we accept that all phylogenetic transformations involve changes occurring during development. The data are thus analogous to those that systematists might come across in nature, but with the following advantage.

Because of our knowledge about the process of development during which the new morphologies were generated, we not only have the full taxonomic map of all possible transformations, but also the "'true phylogeny" which traverses this map (which may be referred to as the "ontogentic tree" to emphasize that it is based on developmental process). This ideal situation is seldom realized in nature, although this has been the goal of recent studies on the relationship between development and evolution with regard to the tetrapod limb by Goodwin & Trainor (1983), Hinchcliffe & Griftiths (1983), Oster et al. (1988) and Shubin & Aiberch (1986); and the extensive analysis of development in the phylogenies of Plethodontids (Wake & Larsen, 1987). As mentioned earlier, the taxonomic map of possible transformations derived from studies of development, is the phenotypic space, or universe of discourse for all hypotheses about evolution and adaptation. We cannot begin to talk about natural selection without a knowledge of the possible forms on which selection may act (Ho & Saunders, 1979; Saunders, 1984; Ho et al., 1986).

Another practical relevance of such a data set is the possible light it may throw on the major assumption behind phylogenetic systematics: i.e. is the orderly process of development that imposes a hierarchical organization on phylogeny which is recoverable by cladistic analyses. The data were coded on the basis of the presence or absence of the 15 characters corresponding to the bifurcation boundaries, but without any assumption of the hierarchical process. Cladograms were generated by the method of strict parsimony in which equal weight is accorded to forward and reverse changes and rooted at the zero state (with no bifurcation boundaries) (see Sokal, 1983). This gave 20 shortest cladograms of length 48, none of which are close to the ontogentic cladogram. A concensus cladogram of the 20 minimal ones consists of 68 steps and comes closest to the ontogenetic cladogram, which has 71 steps (Fig. 5). Major differences persist between the concensus and the ontogenetic cladogram, however; and further analyses using the polymorphism parsimony pro- gramme which precludes character reversals gave no substantial improvement.

These analyses are similar in spirit to those carried out by Sokal (1983) on the "Caminalcules", a set of imaginary organisms with a phylogeny invented by the mathematician Camin. But there is an important distinction. Whereas the phylogeny of the Caminalcules were drawn on purely arbitrary criteria, and hence the recovery

AN E X E R C I S E IN R A T I O N A L T A X O N O M Y 55

~ 0

BI

BICD56 BICD57 (hh)* ~ - -

- - B - - CI23D - - C12D - - CID3 (bit)*

C23DI7 - - - ~ C23D2 (Kr-)*

C23D7 - - . . _ .~ C2D267

C2D15 C2 D2

~ C3D34 C3D4 (kni-)* C (pr-)* D23456 01234 D345 D35 D3 D (wildtype)*

0 A BI

~ B - - C (pr-)*

CI23D - - CI2D - - CID3 (bic)* - - C23DI7 - - C23D2 (Kr-)* - - C23D7 - - C2D267 - - C2D15 - - C2D2

C3D34 - - C3D4 (kni-)*

D23456 - - [)1234

D345 - - D35 - - D3 - - D (wildtype)* _ _ ~ B1C3D

B1CD56 BICD57 (hb)*

Ontogenetic cladogram Consensus parsimony cladogram

__• BICD57 (hb)* CID3 (bic)* C23D2 (Kr-)* C3D4 (kni-)* C (pr-)* D (wildtype)*

Genealogical cladogram

FIG. 5. The ontogenetic cladogram in comparison with the concensus parsimony cladogram and the genealogical cladogram (see text). The mutants are: bic (bicaudal), hb (hunchback), kni (knirps), Kr ( Kripple), pr (paired) (see Akam, 1987).

of the true phylogeny depended on guessing what Camin though to be a "realistic" phylogeny; our phylogeny is based on real development rules, and therefore con- stitutes something closer to a real test o f the ability o f the various numerical procedures to recovering the true hierarchy.

It is o f interest therefore, to consider the reasons why cladistic analysis fails to recover the true ontogenet ic cladogram in its entirety. One which immediately occurs to systematists is that too few characters are available for doing the job: 15 characters being insufficient to resolve 24 taxa. This is a serious problem, but one that is also frequently encountered in real systematics involving morphological characters. Another reason is that the ontogenet ic cladogram has 71 steps and is hence not at all parsimonious. However, the ontogenet ic tree (Fig. 4) with only 32 steps is the most parsimonious o f all. This perhaps, illustrates the failure o f cladistic analysis

56 M.-W. HO

to reflect adequately the actual history (see Mayr, 1981). In this context it is relevant to ask whether cladistic analysis recovers the genealogical relationship between forms.

It so happens that quite a few of the morphologies obtained by exposure to ether vapour also exist as genetic mutants, which enables us to draw a genealogical cladogram. As each mutant is a single step mutation, they are all equidistant from the wildtype, i.e. the), cannot be resolved according to genealogy (see Fig. 5). There is no mystery in this. It is simply because in considering genetic relatedness, we have collapsed the entire development hierarchy. The developmental hierarchy is irrelevant in the hierarchy of genetic relatedness, much as Hennig (1966) has intimated. Thus, if we are to take biological form seriously, we have no option but to investigate the developmental process directly for the purpose of constructing the rational taxonomy of figrms, which is independent of the reconstruction of phylogenetic history of genes.

I am very grateful to Chr is Humphr i e s , Rober t Scot land and Steve Blackmore for s t imula t ing discuss ions and for pe r fo rming the numer ica l analyses. My thanks are also due to Col in Pat terson, David Wake and Brian G o o d w i n for thei r interest and helpful commen t s dur ing the p repa ra t ion of this manuscr ip t .

R E F E R E N C E S

AKAM, M. (1987). The molecular basis for metameric pattern in the Drosophila embryo. Development 101, 1-22.

FOUCAULT, M. (1970). The Order of Things. London: Tavistock Publications. GOLDSCHMJDT, R. (1940). The Material Basis of Evolution. New Haven: Yale University Press. GOODWqN, B. C. & KAUFFMAN, S. A. (1989). Bifurcations, harmonics and the four color wheel model

of Drosophil development. In: Cell to Cell Signalling: From Experiments to Theoretical Models (Goldbeter, A., ed.) London: Academic Press.

GOODWIN, B. C. & TRAINOR, E. H. (1983). The ontogeny and phylogeny of the pentadactyl limb. In: Development and Evolution (Goodwin, B. C., Holder, N. & Wylie, C. C., eds) pp. 75-98. Cambridge: Cambridge University Press.

HINCHLIFFE, J. R. & GRIFFITHS, P. J. (1983). The prechondrogenic patterns in tetrapod limb develop- ment and their phylogenetic significance. In: Development and Evolution (Goodwin, B. C., Holder, N. & Wylie, C. C., eds) pp. 99-122. Cambridge: Cambridge University Press.

HENNIG, W. (1966). Phylogenetic systematics. Urbana, IL: University of Illinois Press. Ho, M. W. (1984). Environment and heredity in development and evolution. In: Beyond neo-Darwinism:

Introduction to the New Evolutionary Paradigm ( Ho, M. W. & Saunders, P. T., eds) pp. 267-288. London: Academic Press.

Ho, M. W. (1986). A structuralism of process. Paper presented at the First Conference of the Osaka Group for the Study of Dynamic Structures, December, 1986, Osaka, Japan. In: Dynamic Structures in Biology (Goodwin, B. C., Sibatani, A. & Webster, G., eds) Edinburgh: Edinburgh University Press. 1989.

Ho, M. W. (1988). How rational can rational morphology be? A post-Darwinian rational taxonomy based on a structuralism of process. Rivista di Biologia 81, 11-55.

Ho, M. W., MATHESON, A., SAUNDERS, P. T., GOODWlN, B. C. & SMAt.LCOMBE, A. (1987). Ether- induced segmentation distrubances in Drosophila melanogaster. Roux's Arch dev. BioL 196, 511-521.

HO, M. W. & SAUNDERS, P. T. (1979). Beyond neo-Darwinism--an epigenetic approach to evolution. J. theor. BioL 78, 573-591.

Ho, M. W., SAUNDERS, P. T. & FOX, S. W. (1986). A new paradigm for evolution. New Sci. 27, 42-43. HULL, D. (1970). Contemporary systematic philosophies. Ann. Rev. EcoL Syst. !, 19-53. KAUFFMAN, S. A., SHYM KO, R. M. & TRABERT, K. (1978). Control of sequential compartment formation

in Drosophila. Science 199, 259-270.

A N E X E R C I S E 1N R A T I O N A L T A X O N O M Y 57

MAYR, E. (1963). Populations, Species, and Evolution. Cambridge, MA: Harvard University Press. MAYR, E. (1981). Biological classification: toward a synthesis of opposing methodologies. Science 213,

510-516. NELSON, G. J. (1978). Ontogeny, phylogeny, paleontology, and the biogenetic law. Svst. Zool. 27, 324-345. NELSON, G. J. (1979). Cladistic analysis and synthesis: principles and definitions, with a historical note

on Adanson ' s Families Des Plantes (1763-1764). Syst. Zool. 28, 1-21. NELSON, G. J. & PLATNIC_'K, N. I. (1984). Systematics and evolution. In: Beyond neo-Darwinism:

Introduction to the New evolutionary Paradigm (Ho, M. W. & Saunders, P. T., eds) pp. 143-158. London: Academic Press.

OSTER, G. F., SHUBIN, N., MURRAY, J. D. & ALBERCH, P. (1988). Evolution and morphogenetic rules: the shape o f the vertebrate limb in ontogeny and phylogeny. Evolution 42, 862-884.

PA'r'rERSON, C. (1983). How does phylogeny differ from ontogeny? In: Development and Evolution (Goodwin, B. C., Holder, N. & Wylie, C. C., eds) pp. 1-31. Cambridge: Cambridge University Press.

PATTERSON, C. (1988). The impact of evolutionary theories on systematics. In: Prospects in Systematics (Hawksworth, D. L., ed.) pp. 59-89. Oxford: Clarendon Press.

SAUNDERS, P. T. (1984). Development and evolution. In: Beyond neo-Darwinism: Introduction to the New Evolutionary Paradigm (Ho, M. W. & Saunders, P. T., eds) London: Academic Press.

SAUNDERS, P. T. & HO, M. W. (1981). On the increase in complexity in evolution. II. The relativity of complexity and the principle of min imum increase. J. theor. Biol. 90, 515-530.

SHUBIN, N. & ALBERCH, P. (1986). A morphogenetic approach to the origin and basic organization of the tetrapod limb. EvoL Biol. 20, 319-387.

SOKAL, R. R. (1983). A phylogenetic analysis of the Caminalcules. If. Estimating the true cladogram. Syst. Zook 32, 185-201.

SOAKAL, R. R. t~ CREVELLO, Z. J. (1970). The biological species concept: a critical evaluation. Am. Nat. 104, 127-53.

v o N BA E R, K. E. (1828). Uber Entwickelungsgeschichte der Thiere. Beobach tung und Reflexion. Konigsberg: Borntrager.

WAKE, D. B. & LARSON, A. (1987). Multidimensional analysis of an evolving lineage. Science 238, 42-48.