The Structure of Evolutionary Theory

by Stephen Jay Gould

  Table 10-1. Evolutionary Similarities in Different Lineages Classified by Two Logical Types (A vs. B or A vs. Not-A) and by Two Criteria (in Realized Structures or in Underlying Generators of the Structures

  A) Logic

  Type 1

  Zone A

  Gray Zone

  Zone B

  Type 2

  Zone A

  Zone Not-A

  B) Name of Phenomenon

  In Type 1

  Homology

  Parallelism

  Convergence

  In Type 2

  Homology

  Homoplasy

  Homoplasy

  C) Basis for Similarity

  In Type 1

  Negative

  constraint

  Positive constraint

  plus independent

  selection

  Independent

  selection

  In Type 2

  Conserved

  descent

  Independent origin

  Independent

  origin

  D) Origin of Similarity for:

  Realized structures

  Conserved

  descent

  Independent

  selection

  Independent

  selection

  Underlying generators of

  the structures

  Conserved

  descent

  Conserved

  descent

  Independent

  selection

  E) Cause of Similarity for Realized Structure

  Inheritance

  Yes

  No

  No

  Current selection

  No

  Yes

  Yes

  F) Cause of Similarity for Underlying Generator

  Inheritance

  Yes

  Yes

  No

  Current selection

  No

  No

  Yes

  [Page 1079]

  level under consideration — see Roth, 1991; Wagner, 1989; Bolker and Raff, 1996. Wing of bats and birds are, after all, convergent as wings, but homologous as forearms).

  At the level of an overt phenotypic structure under explicit consideration, parallelism denies homology and asserts independent origin. But, at the level of the generators for the overt feature — the genes regulating its architecture, and the developmental pathways defining its construction — parallelism af­firms homology as the concept's fundamental meaning and raison d'etre, and the basis for its dichotomous contrast with convergence as alternatives within the more inclusive category of homoplasy. Thus, parallelism does require in­dependent regimes of similar selection, but the resulting phenotypic like­nesses must also be channeled from within by homologous generators.

  (In an odd sense, one might view this old issue of differences between parallelism and convergence as a grand foreshadowing for an important debate that evolutionary biologists have only recently clarified in their minds — but that might have achieved earlier resolution had we all remembered this older discussion: the recognition that cladistic gene-trees do not correspond en­tirely with organism-trees. The capacity for parallelism rests upon organismal branching before gene branching. Continuing the argument, one might also view the first steps in the opposite mode of gene branching before organism branching as a molecular representation of Owen's old concept of serial homology. Paralogs within one organism are serial homologs; different paralogs in two organisms are general homologs; only orthologs in two organisms are special homologs, the heart of the modern concept of pure homology, or Lankester's homogeny — see p. 1071.)

  Framed this way, the maddening complexities and counterclaims of the literature gain immediate clarification. One must then ask why the distinc­tion between parallelism and convergence has bred so much conceptual trou­ble in the past. In particular, the two terms have often been purposefully combined (and demoted) to merely descriptive names for stages in a contin­uum. The terms will then only designate the trivial geometric difference be­tween features evolved independently in two lines that remain at about the same distance in overall phenotype (parallelism) and lineages that become more similar as a consequence of their separate evolution of such functionally comparable features (convergence). One can only wonder, then, why biolo­gists ever bothered to devise explicit terms for mere geometric waystations in a continuum with no interesting causal distinctions. Yet Haas, for example, defended this descriptive and geometric meaning, while his coauthor G. G. Simpson demurred (in Haas and Simpson, 1946). And Willey (1911), in a first book entirely dedicated to the subject (and title) of “Convergence in Evolution,” also denied a meaningful distinction in choosing his single term to encompass the entire subject of separately evolved similarities. Willey wrote (1911, p. xi): “I have used the word convergence in a wide sense ... The [tra­ditional] definitions leave us in the dark as to what degrees of relationship would entitle a given case to be classed as one of parallelism or of conver­gence.” [Page 1080]

  In my judgment, Wake (1991, pp. 543-544) has correctly explained why many biologists blurred the theoretical difference between parallelism and convergence, and then relegated the terms to descriptive waystations in a con­tinuum of results for a single causal process. When the subject of internal constraint faded to a periphery of interest (or even of active denial) within the functionalist orthodoxy of selection's overarching power and adaptation's empirical preeminence at the height of the Modern Synthesis (see Chapter 7 on hardline versions of the Synthesis that peaked in the late 1950's and 1960's), the conceptual distinction of parallelism as a manifestation of inter­nal channeling became uninteresting to most evolutionists (or, in the worst ef­fects of biasing by restrictive theories, even unperceivable). With the defining feature of parallelism thus banished to a limbo of theoretical irrelevance, biologists limited their concern to the support provided for adaptationist pref­erences by the common feature of all homoplasies: the guiding power of inde­pendent selective regimes, whether aided by homologous internal channels (parallelism) or not (convergence), to fashion the same functional result in separate lineages. Wake wrote (1991, pp. 543-544):

  My central theme is the phenomenon of nondivergent evolutionary change among lineages, including convergent morphological evolution, parallelism, and some kinds of reversal — in other words, what phylogeneticists term homoplasy . . . Convergence and parallelism often are considered to constitute strong evidence of the functioning of natural se­lection. Patterson stated, “The general explanation for convergence is functional adaptation to similar environments” (1988, pp. 616-17), but I argue that alternatives must always be considered. In recent years in­creasing attention has been given to the possibility that parallelism is a manifestation of internal design constraints, and so both functionalist and structuralist constructs predict its occurrence.

  As Wake's statement implies, two reasons — one “good” and the other “bad” in the conventional, if simplistic, terms, usually applied to such assess­ments in science — underlie this movement of parallelism to a periphery of limited interest, or to conflation with convergence, a phenomenon of oppo­site theoretical import in judging the differential weights of constraint and adaptation in the origin of homoplastic similarities. Wake correctly identifies the “bad” reason, as an overemphasis on functionalist themes that limited the scope of evolutionary theory during the mid-century's height of enthusiasm for a “hardened” version of the “Modern Synthesis.” Phenomena like paral­lelism, defined by components of internal constraint, did not elicit the atten­tion of many evolutionists during this period.

  But, as Wake recognizes in the last sentence of his statement, parallelism also received limited attention for the eminently “good” reason that, however well defined in a conceptual sense, the crucial distinction between parallel­ism and convergence could not be cashed out in operational terms until recently — for biologists could no
t identify the “homologies of underlying generators” (the shared genetic and developmental bases of independently [Page 1081] evolved structures) needed to distinguish parallelism from the purely adaptational phenomenon of convergence. But evo-devo has become an active field, while the subject of parallelism has been catapulted from a periphery of forced inattention (as a clearly defined but non-operational concept) into the center of evolutionary studies, largely because biologists have now developed criteria for distinguishing the internal constraints of parallelism from the purely selective basis of convergence.

  In short, more than a century after recognizing the important concep­tual distinction, we can finally resolve actual cases by assessing the different contributions to homoplastic similarity made by constraining channels based on homologous generators and directing pathways based on common re­gimes of selection. I shall present the evidence of best cases in the next sec­tion, but will first close this section on conceptual and terminological analysis by citing five chronological episodes in the history of evolutionary debate about parallelism. These linked episodes all exemplify a crucial argument about the importance to general evolutionary theory of current research on the genetics of development: Despite all subsequent confusion and denigra­tion, the concept of parallelism arose as a causal claim about channels of con­straint vs. purely functionalist explanations rooted in natural selection (or some other adaptationist mechanism, as NeoLamarckism remained popular in the early years of this debate) for the evolution of homoplastic resem­blance.

  The interesting literature on parallelism (as opposed to some of the mean­ingless wrangling over terminology) never lost this theoretical context throughout a century of research and commentary. The delay in resolution, and the prolongation of theoretical discussion, did not reflect any lack of clar­ity on the part of evolutionists, especially as explicated by G. G. Simpson, who understood and promoted the concept of parallelism and its potentially radical implications for Darwinian theory. Rather, the persisting frustration about parallelism primarily recorded the inability of geneticists and develop­mental biologists to identify the generators posited as the basis of “latent” or “underlying” homology in the evolution of homoplastic structures deemed parallel rather than convergent. This bolted door of stymied practice has now been unlocked, and we have crossed a threshold into a period of amazingly fruitful research on parallelism in particular, and on the role of developmen­tal constraint based on deep homology in general, for establishing the mark­edly nonrandom clumping of actual organisms within life's potential morphospace.

  The origin of the term “parallelism.” Interestingly, this term first entered evolutionary theory with an entirely different meaning — but for another concept, indeed a far stronger version, of internal channels as major determinants of trends in the history of life: the theory of recapitulation in embryology. In preevolutionary versions, Agassiz had spoken of a “three­fold parallel” of embryological, taxonomic, and paleontological series within larger types. The American paleontologist and evolutionary theorist E. D. Cooe then formalized an evolutionary version of the “law of parallelism” [Page 1082] within recapitulatory theory (see Gould, 1977b, for relevant sources and quotations).

  “The relation of genera,” Cope writes (1887, p. 45), “which are simply steps in one and the same line of development, may be called exact parallel­ism.” In other words, different genera belonging to the same parallel series will run, during their full ontogenies, down varying lengths of a common de­velopmental (and phyletic) trackway. In this sense, the adult of one genus may be virtually identical (exactly parallel in Cope's terms) with the juvenile form of another genus that runs further along the common track during its own ontogeny. Obviously, these common trackways, regulating both the on­togeny and phylogeny of entire series of related genera, invoke a concept of internal constraint with a vengeance. Cope, in this early version of his devel­oping ideas, placed far more stress on internal channeling to explain taxonomic relationships than his later attraction for the functionalist theory of Neo-Lamarckism would allow (see Gould, 1977b and 1981b, for an analysis of Cope's changing views on the relative importance of constraint and func­tion).

  The first use of parallelism in its modern meaning, including its dichotomous pairing with convergence, can also be traced to two of the greatest American vertebrate paleontologists of the late 19th century: W. B. Scott and H. F. Osborn. If the concept can claim a “founding” quotation at all, Scott (1891, p. 362), in a long and famous article on the osteology of early perissodactyls and artiodactyls, invoked degree of taxonomic relationship to distin­guish parallelism from convergence, while emphasizing their common attrib­utes as homoplastic confounders of phylogeny: “But if the various species of the ancestral genus may acquire the new character independently of each other (parallelism), or if the species of widely different genera may gradually assume a common likeness (convergence), then it is plain that such a genus is an artificial assemblage of forms of polyphyletic origin.”

  In his 1895 summer lecture at the Marine Biological Laboratory in Woods Hole, Scott (1896, p. 56) provided a more formal definition: “By parallelism is meant the independent acquisition of similar structure in forms which are themselves nearly related, and by convergence such acquisition in forms which are not closely related, and thus in one or more respects come to be more nearly alike than were their ancestors.”

  More importantly, Scott then explicitly argued that he needed to distin­guish these two categories of homoplasy because parallelism, based on con­straints of inherited channels for preferred change, will generally confound phylogeny less than convergences that arise by similar functional impact upon truly different starting points (1896, p. 58): “It seems the most obvious of commonplaces to say that numerous and close resemblances of structure are prima facie evidences of relationship. Yet the statement is true, even though the resemblances have been independently acquired, because parallel­ism is a more frequently observed phenomenon than convergence, and be­cause the more nearly related any two organisms are, the more likely are they to undergo similar modifications.” [Page 1083]

  Osborn, the patrician “kingmaker” of American paleontology (and quite a potentate in American science in general), cited Scott's definitions in several papers, paying special attention — in the context of his own pluralistic views on the importance of both formal and functional factors as evolutionary causes — to the role of parallelism in combining the push of selection (or some other functionalist cause) with the internal channeling of constraint as the architect of preferred pathways for any agent of “pushing.” For example, in his 1902 paper on “Homoplasy as a law of latent or potential homology,” Osborn had already identified parallelism as falling into a gray zone between the pure analogy of convergence and the pure homology of unaltered in­heritance. With parallelism's notion of “predeterminate variation” (1902, p. 270), Osborn argues, “I think we have to deal with homology or, more strictly, with a principle intermediate between homology and analogy.”

  In a 1905 article on “The ideas and terms of modern philosophical anat­omy,” Osborn then presented a first chart (reproduced here as Fig. 10-13) of relations among these terms, including parallelism and convergence as sub-categories of analogous resemblance (in contrast with homologous resem­blance here restricted to Lankester's notion of homogeny). His chart depicts the geometrical distinction between parallelism and convergence, but his de­finitions follow Scott in relying not on the descriptive difference between par­allel and converging lines, but on “similar characters arising independently in similar or related animals or organs” for parallelism, vs. “similar adaptations arising independently in dissimilar or unrelated animals or organs” for con­vergence.

  These foundational statements indicate both the conceptual clarity and the

  10-13. In his 1905 article, H.F. Osborn treats parallelism and convergence as subcategories of analogous resemblance. But note how he follows Scott in definin
g parallelism by common possession of underlying generating factors, and not by the mere geometry of results.

  [Page 1084]

  nonoperational nature — about the most frustrating situation one can face in science — of the distinction originally made in defining parallelism and convergence. Both Scott and Osborn grasped the importance of separating homoplasy due to underlying homology of generators (“latent or potential homology” in Osborn's apt phrase in the title of his 1902 paper) from homo­plasy rooted exclusively in a similar external context of adaptation. But the biology of their time provided no way to specify or identify these generators. Scott and Osborn therefore had to invoke the entirely unsatisfactory, indirect and vague surrogate of “degree” of taxonomic resemblance — arguing (quite properly of course, however nonoperationally) that the closer the relation­ship between two separated lineages, the more likely that any homoplastic characters will arise by parallelism. Scott expressed his frustration at this unsurmountable situation on the page following his initial definitions (1891, p. 363): “The distinction between the two classes of phenomena [parallelism and convergence] is obviously one of degree rather than of kind, and it will therefore be convenient to consider them together.”