Conrad Hal Waddington (1905–1975) did not respect the traditional boundaries established between genetics, embryology, and evolutionary biology. Rather, he viewed them together as a “diachronic biology.” In this diachronic biology, evolutionary change was caused by heritable alterations in development. Stabilizing selection within the embryo was followed by normative selection on the adult. To explain evolution, Waddington had to invent many concepts and terms, some of which have retained their usage and some of which have not. In this paper I seek to explicate Waddington's ideas and evaluate their usefulness for contemporary evolutionary developmental biology.
In December, 1990, the ASZ Division of the History and Philosophy of Biology history presented a symposium on Development and Macroevolution. During this symposium, Brian Hall (1992) gave an excellent presentation of the legacy of C. H. Waddington's concepts to development and evolution. It is now a decade later, and much has changed. The ASZ is now SICB, and evolutionary developmental biology has its own disciplinary status both in this society and in the worldwide scientific community. New techniques in molecular biology—PCR, in situ hybridization, and computer databases—have made evolutionary developmental biology possible. But can the new wine of molecular biology fit into the conceptual flasks that Waddington had crafted? I would argue that much that molecular biology has shown over the past decade fits remarkably well into the conceptual categories framed by Waddington. In many cases, we have utilized the concept, even if we have jettisoned Waddington's neologism for it. I would like to revisit some of Waddington's concepts and discuss molecular research that is relevant to them. For a review of the research done prior to 1991, see Hall (1992). The concepts I wish to discuss include: I will briefly mention Waddington's description of each of these processes and their relationships to similar theories proposed by other biologists. Then I will discuss experiments and observations made within the past decade that show the relevance of this concept to our attempts to integrate developmental biology, evolutionary biology, and genetics. For further information on the similarities and differences between Waddington's models and similar ones put forth by James Baldwin, C. L. Morgan, H. F. Osborn, and I. I. Schmalhausen, one should see Hall, in preparation).
competence, evocation, and individuation
canalization, chreodes, and homeorhesis
the epigenetic landscape
stabilizing and normative selection
INDUCTION: COMPETENCE, EVOCATION, AND INDIVIDUATION
In the 1920s, Hans Spemann's laboratory initiated a program attempting to find the molecules by which the notochord induced the ectoderm above it to become neural tissue rather than epidermal tissue (see Gilbert and Saxén, 1993, for summary). Waddington dissected the components of neural induction into competence, evocation, and individuation. Competence was the ability to respond to an inducing signal. It was achieved actively and could be selected for. Evocation was the general signal that told a tissue what type of cell it was to become. Individuation was the process by which the cells of that tissue acquired their specific identities. Waddington (1940; see Gilbert 1991a) used these concepts to analyze primary embryonic induction. Here, the gastrula ectoderm had to be made competent to become neural when exposed to the notochord. The notochord produced evocators that told the competent ectoderm to become neural rather than epidermal. Interactions between the neurons and their surroundings would specify which neuron was to be brain and which was to be spinal cord, which was motor neuron and which was sensory neuron. This framework has been confirmed by numerous experiments in both primary and secondary induction. In primary induction, the window of competence to form an organizer is limited to certain stages when the mesoderm can respond to the Nieuwkoop Center signals, and these times correspond to those stages when Xvex-1 gene is active (Shapira et al., 2000). Evocation of the hindbrain and spinal cord neurons is done by the synthesis of BMP inhibitors such as chordin, noggin, and follistatin, but they do not provide any signal for the anterior–posterior or dorsal–ventral specification of those neurons (see Harland and Gerhart, 1997). The individuation of these neurons along these two axes is accomplished separately. The anterior–posterior identities are defined by FGFs and retinoic acid from the posterior, and the dorsal–ventral specification being performed through the Sonic hedgehog and TGF-β paracrine factors provided by the notochord and the epidermis, respectively (Harland and Gerhart, 1997; Liem et al., 1997; Briscoe et al., 1999).
In secondary induction, the evocation of rat kidney mesenchyme by LIF was found to convert the mesenchyme to an epithelium, but only after the mesenchyme had first become competent through the agency of FGF2 and TGF-α (Barasch et al., 1999). The individuation of these epithelial cells into the secretory components of the nephron is done separately and after the evocation. Similarly, Pax6 is found to provide a competence signal for the evocation of the lens. If surface ectoderm from Pax6-deficient rats is combined with wild-type optic cup, lenses will not form. However, if the surface ectoderm is wild-type, and the inducing optic cup is mutant, the ectoderm will form lens tissue (Fujiwara et al., 1994).
Thus, modern research has vindicated Waddington's proposal that induction can be separated into the components of competence, evocation, and individuation. One of the important reasons to see this separation is that regulation can occur separately to each of these components. Thus, we see that BMP4 can act as an inducer on tissues that have different competences. It will tell one cell to divide, another cell to form bone, and another cell to undergo apoptosis.
Waddington's view of the genome was that it was both active and reactive. Unlike most others—both today and then—Waddington did not think of genes solely in terms of gene activity. Rather, he saw the genes in a dialectic of acting and being acted upon. In fact, in Principles of Embryology (1956a) he called his chapter on developmental genetics, ”The Activation of the Genes by the Cytoplasm.” He listed four examples “of the activation, by different types of cytoplasm, of different specifically corresponding genes”: mosaic eggs, induction, chromosome puffs, and Paramecium G-antigens. He viewed the nucleus and the cytoplasm as being in a continual reciprocal dialogue. The epigenotype is the term Waddington (1939) used to capture the idea of the interactions between the genes, gene products, and the environment that led from genotype to phenotype. Today, we might think of the epigenotype as the networks of transcription factors, paracrine factors, and environmental influences that allow the genotype to realize the phenotype.
One modern notion stemming from the epigenotype is that what a given gene does can depend upon its context. In development, the “gene for” phrase has been changed to the “gene involved in” locution. As mentioned above, what BMP4 does depends on its context. Similarly, the enolase protein of the liver, in a different context, becomes a crystallin protein of the lens (Piatigorsky and Wostow, 1991). Even between generations, a gene can differ in its effect. A mutant gene that produces the lack of arms in one person can cause the absence of a thumb in the next generation (Freire-Maia, 1975; Wolf, 1995). In the epigenotype, the gene is not an autonomous entity; it is part of a network of interacting components.
CANALIZATION, CHREODES, AND HOMEORHESIS
Canalization is the property of developmental pathways to produce standard phenotypes despite mild environmental or genetic perturbations. It is the buffering of the system by the epigenotype networks such that most mutations or environmental conditions will not deflect the genotype from realizing the appropriate phenotype of the cell (Waddington, 1940). Canalization allows mutations to build up in the genotype without their being expressed in the phenotype. Thus, it promotes cryptic genetic variation, while preserving the integrity of the differentiating cell. Such genetic variability can be made manifest by changing the environmental conditions and can be selected. (We will discuss this later in the context of genetic assimilation.) Schmalhausen had proposed a similar idea called stabilization (1949; Allen, 1991; Gilbert, 1994).
Canalization necessitates fixed pathways of development. A chreode (Waddington, 1961) is such a developmental trajectory. As, Hall noted, this term has not been accepted into the literature of developmental biology, evolutionary biology, or genetics. Developmental pathway has taken on this meaning. Homeorhesis is the process version of homeostasis. Just as physiological processes are buffered to produce homeostasis, developmental processes are buffered to produce homeorhesis. The integration of genes into buffered and buffering networks produces homeorhesis—the stability of ordered development.
The canalization of development has recently been demonstrated by several independent experiments. Computer simulations (Nijhout and Paulsen, 1997; Nanjundiah, 2000) have shown that the phenotypic effect of variation at a single locus depends critically on the allelic values of other genes in the same pathway and on the frequency of those genes in the population. Moreover, they found that genetic background—the other genes in the genome—is able to buffer so that only a small fraction of the genes that affect the development of a particular trait could be identified in a single sampling.
In addition to the combined effects of background genes, there may be genes that are intimately involved in producing canalyzed phenotypes. Rutherford and Lindquist (1998) showed that a major agent responsible for developmental buffering was the “heat shock protein” Hsp90. Hsp90 is a protein that binds to a set of signal transduction molecules that are inherently unstable. When it binds to them, it stabilizes their tertiary structure so that they can respond to the upstream signaling molecules. Evidence for the role of Hsp90 as a developmental buffer came from mutations of Hsp83, the gene for Hsp90. Homozygous mutations of the Hsp83 are lethal in Drosophila. Heterozygous mutations increase the proportion of developmental abnormalities in the population into which they are introduced. In populations of Drosophila heterozygous for Hsp83, deformed eyes, bristle duplications, and abnormalities of legs and wings appeared. These mutations had been cryptic in the population, and had been masked by the action of Hsp90. When different mutant alleles of Hsp83 were brought together in the same flies, both the incidence and severity of the abnormalities increased. The same abnormalities could be seen when a specific inhibitor of Hsp90 (geldanamycin) was added to the food of wild-type flies, and the types of defects differed between different stocks of flies. The electronic metaphor Rutherford and Lindquist use is that Hsp90 acts as a capacitor for morphological evolution.
Rutherford and Lindquist also provided evidence for the existence of one of Waddington's most controversial ideas—genetic assimilation. Genetic assimilation is the process by which a phenotypic response to the environment becomes, through the process of selection, taken over by the genotype so that it becomes independent of the original environmental inducer. This idea had several predecessors, including those hypotheses of Baldwin, and is essentially the same as Schmalhausen's hypothesis of genetic stabilization. An example used by both Schmalhausen (1949) and Waddington (1942) concerns the calluses on the keels and sterna of ostriches. According to both Schmalhausen and Waddington, the genome of the ostrich has the ability to let the skin form calluses when the skin is abraded. This ability to respond is what is important. If the presence of calluses is adaptive, then that phenotype can be selected such that it forms without abrasion (and appears earlier than the abrasive stimulus). The ability to respond to an external stimulus can be transferred to an internal stimulus.
For genetic assimilation to work, four things have to be shown.
(1) The genome must be responsive to environmental inducers.
(2) The competence to be induced must be transferred from an external inducer to an internal, embryonic inducer.
(3) There has to be cryptic variation within a population so that the physiological induction can be taken over by embryonic inducers.
(4) There must be selection for the phenotype.
The difference between the Waddington–Schmalhausen model of genetic assimilation and the Baldwin Effect is that Baldwin postulated that a single mutation would transfer the inducing signal from the environment to the genotype. For Schmalhausen and Waddington, the transfer of competence came through the cryptic variation already present in the population.
The first tenet of genetic assimilation—that the environment can induce phenotypic variation is now very well established. Dietary polyphenisms, seasonal polyphenisms, and predator-induced polyphenisms are now more familiar to developmental biologists, as life history strategies research has begun to enter developmental biology (Gilbert, 1997; Schlicting and Pigliucci, 1998; Tollrian and Harvell, 1998).
The second tenet of genetic assimilation—that environmental stimuli can become mediated by embryological inducers—is now being shown on a molecular level as well. The evidence is not complete, but it is very interesting. The first comes from the work of Gerd Müller. He has shown that physical stress from embryonic movements can induce several bones within the chick embryo. But more recently he has shown that several bones in the chicken do not form if embryonic movement is suppressed in the chick egg. One of these bones is the fibular crest. This bone connects the tibia to the fibula, and it allows the force of the iliofibularis muscle to pull directly from the femur to the tibia. This direct connection is thought to be important in the evolution of birds (it allows the reduction of the femur), and the fibular crest is a universal feature of the bird hindlimb (Müller and Steicher, 1989). When the bird is prevented from moving within its egg, this bone fails to develop (Wu, 1996; Newman and Müller, 2000). The second is that mechanical stress can form calluses through the same mechanism that normally forms bones. In butterflies, pigmentation differences induced by temperature appear to be genetically fixed in populations experiencing only the extremes of the temperature range (Nijhout, 1991).
The last tenets of genetic assimilation were shown experimentally by numerous researchers including Waddington (1953a, 1956b, 1957), Ho et al., (1983), Matsuda (1982, 1987), and Hall (1992). Waddington, (1953a) for instance, found that in certain strains of wild-type Drosophila melanogaster, heat shock of 40°C during the pupal period caused disruptions in the posterior wing crossvein. Two selection regimens were followed, one where the aberrant flies were bred to one another, and another where the non-aberrant flies were bred to one another. By generation 14, in the crossveinless-selection line, some crossveinless individuals were found even if they did not treat the pupae. More were found in each succeeding generation. A response (and probably not an adaptive one) induced by the environment could be assimilated into the genotype. A similar situation was seen when ether shock caused a phenocopy of the bithorax mutation. Waddington's studies on bithorax phenocopy were confirmed and extended in 1996 by Gibson and Hogness. Ether exposure caused numerous flies to have the bithorax phenotype. Selection procedures then generated flies whose bithorax phenotypes were independent of ether exposure. Using polymerase chain reactions, they found that a large percentage of these phenotypes were due to at least four polymorphisms in the Ultrabithorax gene, the gene that naturally regulates the bithorax phenotype. Moreover, this example of genetic assimilation can be simulated by computer (Behera and Nanjundiah, 1997). Genetic assimilation definitely can occur in the laboratory.
However, the idea that genetic assimilation occurs in nature remained controversial until Rutherford and Lindquist (1998) demonstrated a molecular mechanism for it. The abnormalities that they observed when the Hsp83 was mutated or the Hsp90 protein inactivated did not show simple Mendelian inheritance, but were the outcome of the interactions of several gene products. Selective breeding of the flies with the abnormalities led over a few generations to populations where 80–90% of the progeny had the mutant phenotype. Moreover, these mutants did not keep the Hsp83 mutation. In other words, once the mutation in Hsp83 allowed the cryptic mutants to become expressed, selective matings could retain the abnormal phenotype even in the absence of abnormal Hsp90. Hsp90 appears to be responsible for allowing mutations to accumulate but keeping them from being expressed until the environment changes. Each individual mutation might not change the phenotype, and mating would allow these mutations to be “collected” by members of the population. An environmental change (anything that might stress the cells) would thereby release the hidden phenotypic possibilities of a population. In other words, transient decreases in Hsp90 (resulting from its aiding stress-damaged proteins) would uncover pre-existing genetic interactions that would produce morphological variations. Most of these morphological variations would probably be deleterious, but some might be selected for in the new environment. Such releasing of hidden morphological variation may be responsible for the radiations found in the fossil record.
THE EPIGENETIC LANDSCAPE
This brings us to one of Waddington's central conceits, the epigenetic landscape. The epigenetic landscape (Waddington, 1940, 1962; see Gilbert, 1991b) refers to divergent developmental paths (chreodes) that a cell might take upon finding itself in different conditions. In other words, there are discrete states that a cell could become, and there would not be states in between. A young embryonic cell could become a nerve cell or a skin cell depending on its environment, but it would not become some neuroepidermal cell or any other hybrid. This idea of divergent developmental paths has found a reification in contemporary stem cell biotechnology. Here, human or mouse embryonic stem cells, in vitro representatives of the totipotent inner cell mass blastomeres, are placed into culture. Depending on the paracrine factors placed into the medium, these cells are pushed down different developmental pathways. For instance when mouse embryonic stem cells are exposed to retinoic acid, the stem cells become biased to produce neurons; but when exposed to basic fibroblast growth factor and platelet derived growth factor, the stem cells form glia (Renoncourt et al., 1998; Brüstle et al., 1999).
Waddington also proposed that a mutation or some environmental factor might be strong enough to push a cell over from one chreode into another during development. He used examples such as environmental sex determination and homeosis as possible examples. Not only have these examples been confirmed, but gene knockout experiments have added some others. One of the best examples is that of the Tbx6 gene knockout in mice.
During gastrulation, the dorsal mesoderm is formed, consisting of the dorsalmost notochord (chordamesoderm) and the two parasegmental cords that flank the neural tube. The cells of these paraxial mesodermal cords eventually coalesce into the somites. The Tbx6 protein is of critical importance in specifying the paraxial mesoderm. Tbx6 is thought to encode a DNA-binding transcription factor and it is present in the paraxial mesoderm. Chapman and Papaioannou (1998) constructed gene knockouts wherein the region of the gene encoding the Tbx6 DNA-binding region was interrupted by a neomycin resistance gene. The homozygous Tbx6 mutant mice generated by the gene knockout technique all died in utero. They lacked trunk somites. Rather, their paraxial mesoderm had been transformed into neural epithelia. These mice had three neural tubes running down their back. Moreover, these ectopic neural tubes were patterned appropriately with respect to the notochord and dorsal ectoderm. Thus, the HNF3β message, which is induced in floorplate cells by the notochord, was seen in the region of all three tubes which was closest to the notochord. The expression of the Pax3 gene which is normally seen far from the notochord was also seen far from the notochord in the ectopic paraxial tubes. I think that we can say that the concept of the epigenetic landscape is alive and well in contemporary developmental biology.
COMPONENTS OF EVOLUTION: STABILIZING SELECTION AND NORMATIVE SELECTION
In his 1953(b) essay, “Epigenetics and Evolution,” Waddington analysed the shortcomings of the population genetic account of evolution. He noted (p. 187) that the genetic approach to evolution has culminated in the Modern Synthesis (indeed, this may be the first application of Huxley's book's title to the entire field), but he also noticed that “It has been primarily those biologists with an embryological background who have continued to pose questions…” He names Dalcq, Goldschmidt, and Schmalhausen as critics of the Modern Synthesis. And he now puts forth his own critique. First, the mathematics that give the Synthesis such great prestige had not provided any noteworthy quantitative statements about evolution. Other than Wright's theory of drift, no new insights had come from it.
Moreover, Waddington (1953b) claimed that the Modern Synthesis failed to work in at least three areas. First, much adaptation appeared to be non-genetic and regulated by the environment, not by the inherited genotype. Second, as Goldschmidt had noted, large groups of animals differ from each other in ways not compatible with local races branching off. Accumulations of small mutations in a local group could not separate amphibians from fish or reptiles from amphibians. Waddington noted that Goldschmidt's own hypotheses were so unconvincing to geneticists that they obscured the cogency of Goldschmidt's arguments for these “unbridgeable gaps.” Third, Waddington noted the different rates of evolution seen in the paleontological record.
Waddington (1953b, p. 190) concluded that “It is by paying further attention to the nature of the evolving animal, rather than to that of the environment, that we seem likely to make the most rapid progress in our understanding of evolution.” He claimed that in conventional studies of evolution, the animal is considered either as a genotype (and is studied by geneticists) or as a phenotype (and is studied by taxonomists). What is needed, said Waddington, is an evolutionary study of those processes that get the genotype to the phenotype—the epigenetics of development. Following Goldschmidt, Waddington (p. 190–191) declared, “Changes in genotypes only have ostensible effects in evolution if they bring with them alterations in the epigenetic processes by which phenotypes come into being; the kinds of change possible in the adult form of an animal are limited to the possible alterations in the epigenetic system by which it is produced.”
Waddington (1953b, p. 191) then launched into a critique of the notion of “random mutation,” noting that there are developmental constraints placed on what changes are possible. Therefore, “the consequential changes in the phenotype are not random, since the adult form is produced by the interaction of many genes, and only certain types of alteration of the whole system can be brought about by any conceivable alteration of a single member of the gene complex. No single mutation can produce a pentadactyl limb of vertebrate type on a Drosophila.”
Waddington distinguishes here normalizing selection working on adults and stabilizing selection working during development. Using his epigenetic landscape models, canalization, and genetic assimilation, Waddington shows how both normalizing and stabilizing selection can work together to produce species adapted to particular environments in a manner that can operate over a relatively short time course. (Here, Waddington also points out that the mechanism of genetic assimilation can work either if a single mutant switch gene switched the responsiveness from environment to genetic control or if the switch in competence were due to the actions of several genes already existing in the population.) He shows from his own Drosophila heat shock experiments that such genetic assimilation is real and that traits produced through environmental stimuli can become inherited in the genome. Waddington (1953b, p. 198) concluded by saying that the unbridgeable gaps between large groups of organisms “becomes almost a necessity as soon as we think of development as a cybernetic process, involving stabilization through feed-back and other mechanisms.”
Waddington's notions of normalizing and stabilizing selection have become stabilized in our changing concepts of the roles of genes in evolution (see Gilbert, 1998, 2000). According to the traditional population genetic model, genes were autonomous agents. They acted in adults competing for reproductive success by enabling certain individuals to respond better to their environment than others. These genes were recognized by their differences, hence by their allelic variation, and the differential propagation of these alleles explained natural selection. This is normalizing selection and has been the mainstay of the population genetic model of evolution. The developmental genetic model of evolution follows from Waddington's notions of stabilizing selection. The genes here are not autonomous agents. Rather, they are parts of networks or pathways. The genes of the developmental genetics model of evolution are not structural genes manifest in adults competing for reproductive success. Rather, they are regulatory genes operating during the construction of the embryo. These genes are not discovered through their differences. Rather, they are recognized by their similarities. It is the homology between these genes—Pax6, tinman, Hom-C—that provides evidence for the descent with modification in gene regulation during development.
J. B. S. Haldane (1953), the editor of the volume in which Waddington published this paper, summarized the results using a wonderfully apt developmental metaphor:
“To sum up, then, a number of workers are groping from their own different standpoints towards a new synthesis, while producing facts which do not fit too well into the currently accepted synthesis. The current instar of the evolution theory may be defined by such books as those of Huxley, Simpson, Dobzhansky, Mayr, and Stebbins. We are certainly not ready for a new moult, but signs of new organs are perhaps visible.”
The formulation of the SICB Division of Evolutionary Developmental Biology and the new journals in this area demonstrate that this new developmentally influenced evolutionary theory is indeed in ecdysis and is beginning to spread its wings. The future of evolutionary theory, as Waddington suggested, would be formulating ways to integrate population genetics and developmental genetics through cybernetics. Such programs are currently beginning to forge this integration (Raff et al., 1996; Wagner et al., 1997). This integration is what Waddington (1975) called “Diachronic Biology” an integration of genetics, evolution, and development into a seamless fabric. The Society for Integrative and Comparative Biology might be one of the best places for these approaches to come together.
From the Symposium, Evolutionary Developmental Biology: Paradigms, Problems, and Prospects presented at the Annual Meeting of the Society for Integrative and Comparative Biology, 4–8 January 2000, at Atlanta, Georgia.
Funds for writing this were obtained from the John Simon Guggenheim Foundation.