The Extended Phenotype

by Richard Dawkins

  The same prejudice against explicit mention of genes where one can get away with an individual-level circumlocution is common among biologists. The statement, ‘genes for performing behaviour X are favoured over genes for not performing X’ has a vaguely naive and unprofessional ring to it. What evidence is there for such genes? How dare you conjure up ad hoc genes simply to satisfy your hypothetical convenience! To say ‘individuals that perform X are fitter than individuals that do not perform X’ sounds much more respectable. Even if it is not known to be true, it will probably be accepted as a permissible speculation. But the two sentences are exactly equivalent in meaning. The second one says nothing that the first does not say more clearly. Yet if we recognize this equivalence and talk explicitly about genes ‘for’ adaptations, we run the risk of being accused of ‘genetic determinism’. I hope I have succeeded in showing that this risk results from nothing more than misunderstanding. A sensible and unexceptionable way of thinking about natural selection—‘gene selectionism’—is mistaken for a strong belief about development—‘genetic determinism’. Anyone who thinks clearly about the details of how adaptations come into being is almost bound to think, implicitly if not explicitly, about genes, albeit they may be hypothetical genes. As I shall show in this book, there is much to be said for making the genetic basis of Darwinian functional speculations explicit rather than implicit. It is a good way of avoiding certain tempting errors of reasoning (Lloyd 1979). In doing this we may give the impression, entirely for the wrong reason, of being obsessed with genes and all the mythic baggage that genes carry in the contemporary journalistic consciousness. But determinism, in the sense of an inflexible, tramline-following ontogeny, is, or should be, a thousand miles from our thoughts. Of course, individual sociobiologists may or may not be genetic determinists. They may be Rastafarians, Shakers or Marxists. But their private opinions on genetic determinism, like their private opinions on religion, have nothing to do with the fact that they use the language of ‘genes for behaviour’ when talking about natural selection.

  A large part of this chapter has been based on the assumption that a biologist might wish to speculate on the Darwinian ‘function’ of behaviour patterns. This is not to say that all behaviour patterns necessarily have a Darwinian function. It may be that there is a large class of behaviour patterns which are selectively neutral or deleterious to their performers, and cannot usefully be regarded as the products of natural selection. If so, the arguments of this chapter do not apply to them. But it is legitimate to say ‘I am interested in adaptation. I don’t necessarily think all behaviour patterns are adaptations, but I want to study those behaviour patterns that are adaptations.’ Similarly, to express a preference for studying vertebrates rather than invertebrates does not commit us to the belief that all animals are vertebrates. Given that our field of interest is adaptive behaviour, we cannot talk about the Darwinian evolution of the objects of interest without postulating a genetic basis for them. And to use ‘a gene for X’ as a convenient way of talking about ‘the genetic basis of X’, has been standard practice in population genetics for over half a century.

  The question of how large is the class of behaviour patterns that we can consider to be adaptations is an entirely separate question. It is the subject of the next chapter.

  3 Constraints on Perfection

  In one way or another, this book is largely preoccupied with the logic of Darwinian explanations of function. Bitter experience warns that a biologist who shows a strong interest in functional explanation is likely to be accused, sometimes with a passion that startles those more accustomed to scientific than ideological debate (Lewontin 1977), of believing that all animals are perfectly optimal—accused of being an ‘adaptationist’ (Lewontin 1979a,b; Gould & Lewontin 1979). Adaptationism is defined as ‘that approach to evolutionary studies which assumes without further proof that all aspects of the morphology, physiology and behavior of organisms are adaptive optimal solutions to problems’ (Lewontin 1979b). In the first draft of this chapter I expressed doubts that anyone was truly an adaptationist in the extreme sense, but I have recently found the following quotation from, ironically enough, Lewontin himself: ‘That is the one point which I think all evolutionists are agreed upon, that it is virtually impossible to do a better job than an organism is doing in its own environment’ (Lewontin 1967). Lewontin has since, it seems, travelled his road to Damascus, so it would be unfair to use him as my adaptationist spokesman. Indeed together with Gould he has, in recent years, been one of the most articulate and forceful critics of adaptationism. As my representative adaptationist I take A. J. Cain, who has remained (Cain 1979) consistently true to the views expressed in his trenchant and elegant paper on ‘The perfection of animals’.

  Writing as a taxonomist, Cain (1964) is concerned to attack the traditional dichotomy between ‘functional’ characters, which by implication are not reliable taxonomic indicators, and ‘ancestral’ characters which are. Cain argues forcefully that ancient ‘groundplan’ characters, like the pentadactyl limb of tetrapods and the aquatic phase of amphibians, are there because they are functionally useful, rather than because they are inescapable historical legacies as is often implied. If one of two groups ‘is in any way more primitive than the other, then its primitiveness must in itself be an adaptation to some less specialized mode of life which it can pursue successfully; it cannot be merely a sign of inefficiency’ (p. 57). Cain makes a similar point about so-called trivial characters, criticizing Darwin for being too ready, under the at first sight surprising influence of Richard Owen, to concede functionlessness: ‘No one will suppose that the stripes on the whelp of a lion, or the spots on the young blackbird, are of any use to these animals …’ Darwin’s remark must sound foolhardy today even to the most extreme critic of adaptationism. Indeed, history seems to be on the side of the adaptationists, in the sense that in particular instances they have confounded the scoffers again and again. Cain’s own celebrated work, with Sheppard and their school, on the selection pressures maintaining the banding polymorphism in the snail Cepaea nemoralis may have been partly provoked by the fact that ‘it had been confidently asserted that it could not matter to a snail whether it had one band on its shell or two’ (Cain, p. 48). ‘But perhaps the most remarkable functional interpretation of a “trivial” character is given by Manton’s work on the diplopod Polyxenus, in which she has shown that a character formerly described as an “ornament” (and what could sound more useless?) is almost literally the pivot of the animal’s life’ (Cain, p. 51).

  Adaptationism as a working hypothesis, almost as a faith, has undoubtedly been the inspiration for some outstanding discoveries. von Frisch (1967), in defiance of the prestigious orthodoxy of von Hess, conclusively demonstrated colour vision in fish and in honeybees by controlled experiments. He was driven to undertake those experiments by his refusal to believe that, for example, the colours of flowers were there for no reason, or simply to delight men’s eyes. This is, of course, not evidence for the validity of adaptationist faith. Each question must be tackled afresh, on its merits.

  Wenner (1971) performed a valuable service in questioning von Frisch’s dance language hypothesis, since he provoked J. L. Gould’s (1976) brilliant confirmation of von Frisch’s theory. If Wenner had been more of an adaptationist Gould’s research might never have been done, but Wenner would also not have allowed himself to be so blithely wrong. Any adaptationist, while perhaps conceding that Wenner had usefully exposed lacunae in von Frisch’s original experimental design, would instantly have jumped, with Lindauer (1971), on the fundamental question of why bees dance at all. Wenner never denied that they dance, nor that the dance contained all the information about the direction and distance of food that von Frisch claimed. All he denied was that other bees used the dance information. An adaptationist could not have rested happy with the idea of animals performing such a time-consuming, and above all complex and statistically improbable, activity for nothing. Adaptationism cu
ts both ways, however. I am now delighted that Gould did his clinching experiments, and it is entirely to my discredit that, even in the unlikely event of my having been ingenious enough to think of them, I would have been too adaptationist to have bothered. I just knew Wenner was wrong (Dawkins 1969)!

  Adaptationist thinking, if not blind conviction, has been a valuable stimulator of testable hypotheses in physiology. Barlow’s (1961) recognition of the overwhelming functional need in sensory systems to reduce redundancy in input led him to a uniquely coherent understanding of a variety of facts about sensory physiology. Analogous functional reasoning can be applied to the motor system, and to hierarchical systems of organization generally (Dawkins 1976b; Hailman 1977). Adaptationist conviction cannot tell us about physiological mechanism. Only physiological experiment can do that. But cautious adaptationist reasoning can suggest which of many possible physiological hypotheses are most promising and should be tested first.

  I have tried to show that adaptationism can have virtues as well as faults. But this chapter’s main purpose is to list and classify constraints on perfection, to list the main reasons why the student of adaptation should proceed with caution. Before coming to my list of six constraints on perfection, I should deal with three others that have been proposed, but which I find less persuasive. Taking, first, the modern controversy among biochemical geneticists about ‘neutral mutations’, repeatedly cited in critiques of adaptationism, it is simply irrelevant. If there are neutral mutations in the biochemists’ sense, what this means is that any change in polypeptide structure which they induce has no effect on the enzymatic activity of the protein. This means that the neutral mutation will not change the course of embryonic development, will have no phenotypic effect at all, as a whole-organism biologist would understand phenotypic effect. The biochemical controversy over neutralism is concerned with the interesting and important question of whether all gene substitutions have phenotypic effects. The adaptationism controversy is quite different. It is concerned with whether, given that we are dealing with a phenotypic effect big enough to see and ask questions about, we should assume that it is the product of natural selection. The biochemist’s ‘neutral mutations’ are more than neutral. As far as those of us who look at gross morphology, physiology and behaviour are concerned, they are not mutations at all. It was in this spirit that Maynard Smith (1976b) wrote: ‘I interpret “rate of evolution” as a rate of adaptive change. In this sense, the substitution of a neutral allele would not constitute evolution …’ If a whole-organism biologist sees a genetically determined difference among phenotypes, he already knows he cannot be dealing with neutrality in the sense of the modern controversy among biochemical geneticists.

  He might, nevertheless, be dealing with a neutral character in the sense of an earlier controversy (Fisher & Ford 1950; Wright 1951). A genetic difference could show itself at the phenotypic level, yet still be selectively neutral. But mathematical calculations such as those of Fisher (1930b) and Haldane (1932a) show how unreliable human subjective judgement can be on the ‘obviously trivial’ nature of some biological characters. Haldane, for example, showed that, with plausible assumptions about a typical population, a selection pressure as weak as 1 in 1000 would take only a few thousand generations to push an initially rare mutation to fixation, a small time by geological standards. It appears that, in the controversy referred to above, Wright was misunderstood (see below). Wright (1980) was embarrassed at finding the idea of evolution of nonadaptive characters by genetic drift labelled the ‘Sewall Wright effect’, ‘not only because others had previously advanced the same idea, but because I myself had strongly rejected it from the first (1929), stating that pure random drift leads “inevitably to degeneration and extinction”’. I have attributed apparent nonadaptive taxonomic differences to pleiotropy, where not merely ignorance of an adaptive significance.’ Wright was in fact showing how a subtle mixture of drift and selection can produce adaptations superior to the products of selection alone (see pp. 39–40).

  A second suggested constraint on perfection concerns allometry (Huxley 1932): ‘In cervine deer, antler size increases more than proportionately to body size … so that larger deer have more than proportionately large antlers. It is then unnecessary to give a specifically adaptive reason for the extremely large antlers of large deer’ (Lewontin 1979b). Well, Lewontin has a point here, but I would prefer to rephrase it. As it stands it suggests that the allometric constant is constant in a God-given immutable sense. But constants on one time scale can be variables on another. The allometric constant is a parameter of embryonic development. Like any other such parameter it may be subject to genetic variation and therefore it may change over evolutionary time (Clutton-Brock & Harvey 1979). Lewontin’s remark turns out to be analogous to the following: all primates have teeth; this is just a plain fact about primates, and it is therefore unnecessary to give a specifically adaptive reason for the presence of teeth in primates. What he probably meant to say is something like the following.

  Deer have evolved a developmental mechanism such that growth of antlers relative to body size is allometric with a particular constant of allometry. Very probably the evolution of this allometric system of development occurred under the influence of selection pressures having nothing to do with the social function of antlers: probably it was conveniently compatible with pre-existing developmental processes in a way which we shall not understand until we know more about the biochemical and cellular details of embryology. Maybe ethological consequences of the extra large antlers of large deer exert a selective effect, but this selection pressure is likely to be swamped in importance by other selection pressures concerned with concealed internal embryological details.

  Williams (1966, p. 16) invoked allometry in the service of a speculation about the selection pressures leading to increased brain size in man. He suggested that the prime focus of selection was on early teachability, at an elementary level, of children. ‘The resulting selection for acquiring verbal facility as early as possible might have produced, as an allometric effect on cerebral development, populations in which an occasional Leonardo might arise.’ Williams, however, did not see allometry as a weapon against the use of adaptive explanations. One feels that he was rightly less loyal to his particular theory of cerebral hypertrophy than to the general principle enunciated in his concluding rhetorical question: ‘Is it not reasonable to anticipate that our understanding of the human mind would be aided greatly by knowing the purpose for which it was designed?’

  What has been said of allometry applies also to pleiotropy, the possession by one gene of more than one phenotypic effect. This is the third of the suggested constraints on perfection that I want to get out of the way before embarking on my main list. It has already been mentioned in my quotation from Wright. A possible source of confusion here is that pleiotropy has been used as a weapon by both sides in this debate, if indeed it is a real debate. Fisher (1930b) reasoned that it was unlikely that any one of a gene’s phenotypic effects was neutral, so how much more unlikely was it that all of a gene’s pleiotropic effects could be neutral. Lewontin (1979b), on the other hand, remarked that ‘many changes in characters are the result of pleiotropic gene action, rather than the direct result of selection on the character itself. The yellow color of the Malpighian tubules of an insect cannot itself be the subject of natural selection since that color can never be seen by any organism. Rather it is the pleiotropic consequence of red eye pigment metabolism, which may be adaptive.’ There is no real disagreement here. Fisher was talking of the selective effects on a genetic mutation, Lewontin of selective effects on a phenotypic character; it is the same distinction, indeed, as I was making in discussing neutrality in the biochemical geneticists’ sense.

  Lewontin’s point about pleiotropy is related to another one which I shall come on to below, about the problem of defining what he calls the natural ‘suture lines’, the ‘phenotypic units’ of evolution. Sometimes the dual effects of a g
ene are in principle inseparable; they are different views of the same thing, just as Everest used to have two names depending on which side it was seen from. What a biochemist sees as an oxygen-carrying molecule may be seen by an ethologist as red coloration. But there is a more interesting kind of pleiotropy in which the two phenotypic effects of a mutation are separable. The phenotypic effect of any gene (versus its alleles) is not a property of the gene alone, but also of the embryological context in which it acts. This allows abundant opportunities for the phenotypic effects of one mutation to be modified by others, and is the basis of such respected ideas as Fisher’s (1930a) theory of the evolution of dominance, the Medawar (1952) and Williams (1957) theories of senescence, and Hamilton’s (1967) theory of Y-chromosome inertness. In the present connection, if a mutation has one beneficial effect and one harmful one, there is no reason why selection should not favour modifier genes that detach the two phenotypic effects, or that reduce the harmful effect while enhancing the beneficial one. As in the case of allometry, Lewontin took too static a view of gene action, treating pleiotropy as if it was a property of the gene rather than of the interaction between the gene and its (modifiable) embryological context.