The Structure of Evolutionary Theory

by Stephen Jay Gould

  Parallelism will only define an interesting antithesis to convergence if the underlying homology prescribes a highly distinctive, detailed, and strongly determinative channel of constraint — for only then will the homoplastic re­sult owe its primary form to the structure of the internal channel, and not to the functional processes of adaptation acting from outside. But if, on the other hand, the underlying homology only generates a simple immediate product, leading to a broad and non-specific range of potential outcomes, the homology establishes no meaningful channel of internal constraint, and makes no contribution to the revisionary power of this theme within evolu­tionary theory — that is, to move away from an endpoint of pure Darwinian functionalism towards a more comprehensive theory, enriched by contrasting perspectives based on structural principles of internal channeling.

  In the light of our burgeoning knowledge of genetic sequences and their ac­tions, homology of some sort or level will always be found in underlying gen­erators of similar end products — if only (however much the example becomes a reductio ad absurdum) because all organisms share the same genetic code by common ancestry. But no one would argue that we should redescribe a classical case of convergence as parallelism simply because the markedly dif­ferent developmental pathways of the two adaptations both rest upon the ac­tion of genes made of DNA!

  The analogy in the title of this subsection may help to clarify the central [Page 1136] issue of this important discussion. We need to develop criteria for ordering and evaluating our highly varied and ever-growing compendium of homoplastic results generated along homologous developmental pathways — for these cases fall along a continuum from narrow and controlling channels of constraint to insignificant sharing of nonspecific building blocks. When Pha­raoh “made the children of Israel serve with rigor” (Exodus 1:13), they fabri­cated bricks to use in a full range of buildings: “And they built for Pharaoh treasure cities, Pithom and Raamses” (Exodus 1:11). Now if these bricks built every structure in the city, from great pyramids to public toilets, we might identify a homologous generator of all final products (bricks of the same composition made by the same people in the same way over a continu­ous stretch of time). But we could scarcely argue that these homologous gen­erators exerted any important constraint over the differing forms of Pha­raoh's final products — if only because all realized architectural diversity shared the same building blocks.

  But if I note a majestic portico of Corinthian columns in front of a building in modern Manhattan, I recognize a strong internal constraint imposed by an architectural module of very different status. The Corinthian column, last and most ornate of the classical orders, consists of a slender fluted shaft (with 24 flutes in “standard” examples), capped by a striking, distinctive, and elaborate capital (the defining “species” character in a taxonomic analogy) adorned with stylized acanthus leaves. Few Greek examples survive, but the Romans then used Corinthian columns extensively and for several centuries. Vetruvius, who wrote the only surviving work on classical architecture, de­scribed such columns in detail in the 1st century bc, and later builders of the Italian Renaissance replicated the design in all aspects of form and propor­tion, whence, ever since, buildings in classical style have often used Corinthian columns on their facades, porticos and lobbies.

  Like Pharaonic bricks, Corinthian columns hold clear status as homolo­gous underlying generators for their continuous phyletic history and stable form. But whereas Pharaonic bricks did little to constrain a resulting building by their form or structural character, and would not therefore sustain an in­teresting interpretation of parallelism for two similar buildings that happened to employ them in construction (if only because many other, very unsimilar, buildings in town also use the same bricks), Corinthian columns do exert a strong structural constraint from an inherited past (a homology) that can help us to identify and distinguish buildings, even 2500 years after the inven­tion of this unchanged form.

  First of all, we can identify the lineage of the element just by looking (for the acanthus leaves mark the species), so I know the source of constraint before proceeding any further, whereas a brick may be difficult to peg as Pharaonic, or as a product of independent invention for a simple, obvious and utilitarian form (a Chinese version, for example). Second, tradition and intrinsic form dictate that this large and elaborate column only be used in limited ways (whereas my Pharaonic brick can build almost anything) — so the choice of the column constrains the form and function of the building. [Page 1137] For example, in commercial areas of Manhattan, we can be pretty sure that a set of Corinthian columns will be fronting a bank, a government building or a church — so the generator constrains the function. In residential areas, we can be confident that a similar set of columns will mark the domicile of wealthy folks, not the entrance to a public housing project — so the generator also con­strains the location and social setting.

  In short, and emphasizing the evolutionary analogy, two similar build­ings made of identical bricks in the cities of Pithom or Raamses are not constrained to be alike by the structural properties of their admittedly ho­mologous building blocks. The bricks represent a lowest level, non-specific, non-constraining homologous generator, and the similarity between the two buildings is convergent, a result of architectural decisions about good form for an intended purpose — that is, a product of external selection based on re­quired function, not of internal channeling imposed by component parts. But the similar form of a town hall in modern America and a market hall in an­cient Rome, as highlighted by their nearly identical facades of Corinthian col­umns, must be attributed, in large part, to the complex, highly specific, phyletically stable design of these chosen architectural modules, which therefore do constrain the form and function of the buildings in important ways. We may therefore ascribe much of the similarity to parallelism, based on the com­mon choice of a homologous building element that establishes a channel of expectation, and has done so for millennia (and also includes too much com­plexity and too little flexibility to be used in many ways beyond the tradi­tional employment).*

  Thus, examples of homologous underlying generators form a continuum from Pharaonic bricks, which are too simple, general, and multipurpose to constrain a final result in important ways, to Corinthian columns, which are sufficiently complex, structurally limited in potential utility, and restricted by a long and stable history of traditional employment, to channel any building into just a few recognized forms and functions. When underlying homolo­gous generators operate like Corinthian columns, they entail interpretations of parallelism, rather than pure convergence, for structural and functional similarities in resulting adult anatomies. But when homologous generators operate like Pharaonic bricks, they usually do not strongly constrain similari­ties between two independent structures built with their aid, and we would [Page 1138] not automatically, on this basis, ascribe such similarities to parallel evolution. I do not know how to ordain hard and fast rules for breaking this smooth continuum into sharp domains of bricks that permit interpretations of con­vergence vs. columns that imply parallelism — but I trust that the analogy will clarify the issues involved, which must then be adjudicated on a case-by-case basis.

  Heretofore, as the argument of this chapter demanded, I have been present­ing cases of biological equivalents to Corinthian columns, leading to reassignments of convergence to parallelism (although I did raise the “brick” issue in wondering whether the signaling system behind dorsoventral inversion of ar­thropods and vertebrates might be too broad to bear Geoffroy's interpreta­tion — because such a general system may regulate many other distinctions as well, and may therefore become prone to independent cooptation (in differ­ent form) by two separate groups. Therefore, the facts of DV inversion do not yet guarantee an explanation as two different specializations of a homolo­gous and archetypal ancestral state — see p. 1122).

  But many examples of homologous generators acting more like bricks than columns have also been accumulat
ing in the literature. Such examples imply interpretations more favorable to adaptation (and convergence) than to con­straint (and parallelism) for two distinct reasons: first, because bricks are too general and non-specific in their operation to exert much constraint upon the complex form of a final product; and, second, because bricks are sufficiently simple and multifarious in their range of potential developmental utility that each of two lineages now using the same brick in the same way, may have co-opted this architectural module independently, and from a different ancestral use in each case. In this second circumstance, the functional similarity of bricks in the two lineages would not even be homologous, given their inde­pendent cooptation from different sources, although the bricks remain ho­mologous in genetic structure (by attribution of the requisite similarity in nucleotide sequences to a more distant common ancestor).

  To cite just two examples of bricks (that is, very general and effectively nonconstraining homologies of genetic and developmental architecture) from the recently published genome of the nematode, C. elegans: First, for homologies going “down” life's traditional ladder, Chalfie (1998, p. 620) noted the brick-like nature of worm genes with homologs in yeast: “Most orthologues [to yeast] in the worm are needed for . . . core functions, such as intermediary metabolism, DNA-, RNA- and protein-metabolism, transport and secretion, and cytoskeletal structure. In contrast, yeast has no ortho­logues for many of the proteins involved in intercellular signaling and gene regulation in C. elegans.” Second, for brick-like homologies going “up” the same fallacious ladder, Bohm et al. (1997) found that the par-1 gene of C. elegans, which codes for a protein that activates the markedly asymmetrical division of cells in the first embryonic cleavage, has a mammalian homolog that regulates the polarization of epithelial cells.

  This central issue of Pharaonic bricks and Corinthian columns has become most salient in the fascinating and rapidly developing literature on the extent [Page 1139] and meaning of similarities in development between the appendages of arthropods and vertebrates (Tickle, 1992; Tabin, Carroll and Panganiban, 1999; Panganiban et al., 1997; Shubin, Tabin and Carroll, 1997; Arthur, Jewett and Panchen, 1999; Minelli, 2000; for example). I will state my own tentative reading of these preliminary data up front: most documented homologies are too brick-like to impart a sufficiently strong and specific con­straint for validating either the actual homology of limbs themselves, or even a claim for predominant parallelism in the evolution of homoplastic append­ages. This situation may be contrasted with the highly and specifically chan­neled developmental homologies underlying the establishment and differenti­ation of major body axes, several aspects of segmentation itself, and the evolution of important homoplastic organs at several levels, including eyes among phyla and maxillipeds among crustacean taxa. These homologies are more than sufficiently column-like to validate channels of internal constraint as primary determinants of specific final products. However, some attributes of homoplastic features in arthropod and vertebrate appendages do offer in­triguing hints that, even here, developmental homologies may be sufficiently column-like in some cases to implicate constraints of internal generating channels as major causes of similarity in adult structures.

  As a prime example of a brick-like developmental homology, now regarded as too broad and loosely constraining to specify important details of final products as outcomes shaped by internal channels, but often regarded as more column-like in the first excitement of discovery, the Drosophila distal-less gene (Dll) is expressed at the distal tip of developing appendages and seems important in regulating their outgrowth from the body axis (Cohen et al., 1989). In the mid 1990's, researchers found a mammalian homolog (called Dlx) that seems to operate in virtually the same way, with expression along the distal edge of the chick wing bud (Carroll et al., 1994; Panganiban et al., 1995).

  But as studies proceeded, an embarras de richesses soon became appar­ent, as distal-less homologs were found at the terminal regions of almost any structure that grows out from a central mass or body axis in all three great groups of bilaterians — including annelid parapodia, onychophoran lobopodia, tunicate ampullae and echinoderm tube feet (Panganiban et al., 1997). Lee and Jacobs (1999) then pointed out that not only does distal-less seem to regulate the proximodistal axis of any outgrowth, but it also tends to show preferred action in early embryos (including maternal transcripts in several cases), in animal poles and anterior regions of developing embryos, and in ectodermal germ layers. Thus, distal-less may not only display the broad function of regulating outgrowths at their distal tips; it may also oper­ate in the service of even more basic distinctions that can only be designated as early, anterior and top. Distal-less, in this sense, must be regarded as a quintessential Pharaonic brick of protean character, or just about as non-specifically unconstraining as an internal developmental element can be. If anyone wanted to argue that insect and vertebrate appendages should be deemed homologous because both are regulated by distal-less homologs, then [Page 1140] these structures are only homologous as outgrowths in all Bilateria, and the claim becomes almost as meaninglessly broad as saying that I am homolo­gous to each of my E. coli residents because we are both made of DNA inher­ited from a common ancestor. Moreover, with such a broad range of func­tions, and such ubiquity of occurrence, distal-less genes might have been independently coopted from different copies with different utilities, rather than commonly employed from the same ancestral source, in the arthropod and vertebrate forebears that first used them to regulate the outgrowth of ap­pendages.

  Panganiban et al. (1997, p. 5165) state the case for such a broad and rela­tively unconstraining homology: “The most straightforward explanation for these observations is that the last common ancestor of the protostomes and deuterostomes had some primitive type of body wall outgrowths, e.g., a sen­sory or perhaps a simple locomotory appendage, and that the genetic cir­cuitry governing the outgrowth of this structure was deployed at new sites many times during evolution.” Shubin et al. (1997, p. 647) then add a reason­able, but admittedly indecisive, argument for favoring common ancestry over independent cooptation: “The expression of Dll-related genes could repre­sent convergent utilization of the gene. However, the fact that out of the hun­dreds of transcription factors that potentially could have been used, Dll is ex­pressed in the distal portions of appendages in six coelomate phyla makes it more likely that Dll was already involved in regulating body wall outgrowths in a common ancestor of these taxa.”

  On the other hand, when homologies of underlying generators (for homo-plastic structures between phyla) begin to involve several genes and their complex interactions — rather than just one product expressed at the distal tip of any outgrowth — then the homology attains sufficient definition and spe­cificity to act as a constraining Corinthian column of positive evolution­ary channeling, rather than as an all-purpose Pharaonic brick for building nearly any kind of structure that natural selection might favor. For exam­ple, no one would argue that the chick forearm and fly wing are homologous as flight appendages, if only for the obvious and compelling reason that basal chordates — not only as inferred from living surrogates, but also as rea­sonably well represented in the fossil record of the Cambrian explosion and its sequelae — lack paired appendages entirely. But accumulating evidence now indicates that all three major axes (anteroposterior, dorsoventral, and proximodistal) may be established (or at least strongly regulated) by homo­logous, and respectably complex, genes and their interactions — a strong case for meaningful column-like constraint in this important anatomical system as well.

  In the wing or leg imaginal disc of Drosophila, hedgehog acts to establish the AP axis by initial expression in the posterior compartment of the disc. In response to Hedgehog, a thin layer of cells along the border of both anterior and posterior compartments produces another protein encoded by the dpp gene, “dpp, in turn, is a long-range signal providing positional information, and hence differential AP fates, to cells in both compartments” (Sh
ubin et al., [Page 1141] 1997, p. 645). The vertebrate limb develops in a similar way under homolo­gous influences. Sonic Hedgehog (Shh), a homolog of Drosophila hedgehog, is also located in the posterior part of the limb bud, and also induces a homolog of dpp, called Bmp-2 in its vertebrate version, which then acts in differentiating the AP axis. Interestingly, and in a striking confirmation of im­portant homology in function as well as structure, misexpression of either Shh or hedgehog on the anterior rather than the posterior edge of the devel­oping appendage causes the same striking malformation: mirror image dupli­cations at the anterior border.

  In organizing the proximodistal structure of the Drosophila wing, a spe­cialized set of cells, called the wing margin, runs along the DV border of the imaginal wing disc. The fringe gene establishes the edge of the developing wing at the interface between cells that do and do not express fringe. The ver­tebrate apical ectodermal ridge (AER), like the Drosophila wing margin, also runs along the DV border of the developing limb bud. Radical fringe, a verte­brate homolog of fringe, is also expressed in the dorsal region limb ectoderm before the AER forms. Moreover, at the border of cells that do and do not ex­press radical fringe, Ser-2, a homologue of Drosophila serrate (an important element in the downstream cascade regulated by fringe) becomes expressed, and the AER then forms.

  Data for the DV axis are less well developed, but “genes specifying DV po­larity in both groups have been identified” (Shubin et al., 1997, p. 646). For example, in Drosophila, the expression of apterous helps to define the dorsal compartment of the wing disc and also specifies dorsal cell fates. A related, but not clearly homologous vertebrate gene, Lmx-1, defines a dorsal com­partment of the vertebrate limb, and also conveys dorsal cell fate.

  Shubin et al. (1997, p. 646) summarize the evolutionary meaning of these complex, column-like developmental constraints: “The simplest phylogenetic implication to draw from these comparisons is that individual genes that are expressed in the three orthogonal axes are more ancient than either insect or vertebrate limbs ... either similar genetic circuits were convergently recruited to make the limbs of different taxa or a set of these signaling and regulatory systems are ancient and patterned a structure in the common ancestor of protostomes and deuterostomes.”