The Extended Phenotype

by Richard Dawkins

  Sargent’s (1969a) intuition differs from Kettlewell’s. He doubts the self-inspection theory, and thinks that the two morphs of B. betularia differ genetically in their background preference. He has no evidence about B. betularia itself, but has done some ingenious experiments on other species. For example, he took members of a dark species and of a light species, and painted the circum-ocular hairs in an attempt to ‘fool’ the moths into choosing a background that matched the painted hairs. They obstinately persisted in choosing a background that matched their genetically determined colour (Sargent 1968). Unfortunately, however, this interesting result was obtained with two different species, not with dark and light morphs of one species.

  In another experiment which was done with a dimorphic species, Phigalia titea, Sargent (1969a) simply failed to get the result Kettlewell had with B. betularia. The P. titea individuals, whether the dark or the light morph, chose to sit on light backgrounds, presumably the appropriate background for the light ancestral form of the species. What is needed is for somebody to repeat Sargent’s key experiment of painting the parts of a moth’s body that it can see, but using a dimorphic species like B. betularia which is known to show morph-specific background choice. Kettlewell’s theory would predict that moths painted black will select black backgrounds, and moths painted light will select light backgrounds, regardless of whether they are genetically carbonaria or typica. A purely genetic theory would predict that carbonaria will choose dark backgrounds and typica light backgrounds, regardless of how they are painted.

  If the latter theory turned out to be right, would this result give aid and comfort to the green-beard theory? A little, perhaps, since it would suggest that a morphological character and behavioural recognition of something resembling that morphological character can be, or can rapidly become, closely linked genetically. It must be remembered, here, that in the moth crypsis example there is no suggestion that we are dealing with an outlaw effect. If there are two genes, one controlling colour and the other background choice behaviour, both stand to gain from the presence of the appropriate other one, and neither of the two is in any sense an outlaw. If the two genes started out distantly linked, selection would favour closer and closer linkage. It is not clear that selection would similarly favour the close linkage of a ‘green-beard gene’ and a gene for green-beard recognition. It seems that the association between the effects would have to be there by luck right from the start.

  The green-beard effect is all about one selfish gene looking after copies of itself in other individuals, regardless of, indeed in spite of, the probability that those individuals will share genes in general. The green-beard gene ‘spots’ copies of itself, and it thereby seems to work against the interests of the rest of the genome. It appears to be an outlaw in the sense that it makes individuals work and pay costs, for the benefit of other individuals who are not particularly likely to share genes other than the outlaw itself. This is the reason Alexander and Borgia (1978) call it an outlaw, and it is one of the reasons for their scepticism about the existence of green-beard genes.

  But it is actually not obvious that green-beard genes would, if they ever arose, be outlaws. The following caution is urged by Ridley and Grafen (1981). Our definition of an outlaw makes reference to its provoking of modifiers at other loci which tend to suppress its phenotypic effects. At first sight it seems clear that green-beard genes would indeed provoke suppressing modifiers, because the modifiers would not tend to have copies in the bodies of the (unrelated) green-bearded individuals cared for. But we must not forget that a modifier, if it is to have any effect on the phenotypic expression of a green-beard gene, is likely itself to be in a green-bearded body, and is therefore in a position to benefit from receiving altruism from other green-bearded individuals. Moreover, since those other green-bearded donors are not particularly likely to be relatives, the costs of their altruism will not be felt by copies of the would-be modifiers. A case could therefore be made that would-be modifiers at other loci will gain rather than lose from sharing a body with a green-beard gene. To this it cannot be objected that the costs of paying out altruism to other green-bearded individuals may outweigh the benefits of receiving altruism from other green-bearded individuals: if this were true there would be no question of the green-beard gene spreading in the first place. The essence of Ridley and Grafen’s point is that if (which is unlikely) the green-beard gene has what it takes to spread through the population at all, the costs and benefits of the situation will be such as to favour modifiers that enhance rather than reduce the effect.

  In evaluating this point, everything depends on the exact nature of what we are calling the green-beard phenotype. If the entire pleiotropic dual phenotype—green beard plus altruism towards green-bearded individuals—is regarded as a package deal, which modifiers can suppress or enhance only as a unit, then Ridley and Grafen are surely right that green-beard genes are not outlaws. But, of course, as they themselves stress, a modifier that could detach the two phenotypic effects from each other, suppressing the altruistic phenotype of the green-beard gene while not suppressing the green beard itself would certainly be favoured. A third possibility is the special case of a green-beard gene that caused parents to discriminate in favour of those of their children that happened to share the recognition character. Such a gene would be analogous to a meiotic-drive gene, and would be a true outlaw.

  Whatever we may feel about Ridley and Grafen’s point about the green-beard effect, it is clear that genes mediating altruism towards close kin, and favoured by conventional kin-selection pressure, are definitely not outlaws. All the genes in the genome have the same statistical odds of gaining from the kin altruism behaviour, for all have the same statistical odds of being possessed by the individual benefited. A ‘kin-selection gene’ is, in a sense, working for itself alone, but it benefits the other genes in its genome as well. There will therefore not be selection in favour of modifiers that suppress it. Armpit self-inspection genes would be a special case of kin-recognition genes, and are likewise not outlaws.

  I have been negative about the plausibility of the green-beard effect. The postulated favouritism based on sex chromosomes, which I have already mentioned, is a special case of the green-beard effect, and is perhaps the least implausible one. I discussed it in the context of within-family favouritism: elder siblings were supposed to discriminate among their younger siblings on the basis of probability of sharing sex chromosomes, sex itself being used as the label (‘green beard’). This is not too wildly improbable because, if Y chromosomes do not cross over, instead of having to postulate a single pleiotropic green-beard ‘gene’ we can postulate a whole ‘green-beard chromosome’. It is sufficient that the genetic basis for sexual favouritism should occur anywhere on the sex chromosome concerned. One might make a similar argument for any substantial portion of a chromosome which, say because of inversion, did not cross over. It is conceivable, therefore, that a true green-beard effect of some kind might one day be discovered.

  I suspect that all present examples of what might appear to be green-beard effects are, in fact, versions of the armpit self-inspection effect. Thus Wu et al. (1980) placed individual monkeys, Macaca nemestrina, in a choice apparatus where they had to choose to sit next to one of two offered companions. In each case one of the two companions offered was a half sibling, related through the father but not the mother; the other was an unrelated control. The result was a statistically significant tendency for individuals to choose to sit next to half siblings rather than unrelated controls. Note that the half siblings concerned were not related in the maternal line: this means that there is no possibility of their recognizing an odour acquired from the mother, say. Whatever the monkeys are recognizing, it comes from the shared father, and this suggests, in some sense, recognition of shared genes. My bet is that the monkeys recognize resemblances of relatives to perceived features of themselves. Wu et al. are of the same opinion.

  Greenberg (1979) studied the primitivel
y social sweat bee Lasioglossum zephyrum. (Seger, 1980, refers to this work under the picturesque heading, ‘Do bees have green setae?’.) Where Wu et al. used choice of sitting partner as a behavioural assay, Greenberg used the decision of a guard worker whether to admit or exclude another worker seeking entrance to the nest. He plotted the probability of a worker’s being admitted against her coefficient of relationship with the sentinel. Not only was there an excellent positive correlation: the slope of the line was almost exactly one, so the probability of a sentinel admitting a stranger was approximately equal to the coefficient of relationship! Greenberg’s evidence convinced him that ‘The genetic component is therefore in odor production and apparently not in the perceptual system’ (p. 1096). In my terminology, Greenberg’s words amount to the statement that he is dealing with the armpit effect, not with the green-beard effect. Of course the bees may, as Greenberg believes, have inspected relatives with whom they were already familiar rather than their own ‘armpits’ (Hölldobler & Michener 1980). It is still essentially an example of the armpit rather than the green-beard effect, in which case there is no question of the genes responsible being outlaws. A particularly elegant study which comes to a similar conclusion is that of Linsenmair (1972) on the family-specific chemical ‘badge’ of the social desert woodlouse Hemilepistus reaumuri. Similarly, Bateson (1983) provides intriguing evidence that Japanese quail discriminate their first cousins from their siblings and from more distant relatives, using learned visual cues.

  Waldman and Adler (1979) have investigated whether tadpoles preferentially associate with siblings. Colour-marked tadpoles taken from two clutches were allowed to swim freely around a tank; then a grid was lowered into the tank, trapping each tadpole in one of sixteen compartments. There was a statistically significant tendency for tadpoles to end up closer to siblings than to non-siblings. Perhaps unfortunately, the experimental design does not rule out a possible confounding effect of genetically determined ‘habitat selection’. If there was a genetically determined tendency to, say, hug the edges of the tank rather than the middle, genetic relatives might as a consequence be expected to end up in the same parts of the tank. The experiment does not, therefore, unequivocally demonstrate recognition of relatives or a preference for associating with relatives as such, but for many theoretical purposes this does not matter. The authors introduced their paper with reference to Fisher’s (1930a) kin selection theory of the evolution of aposematism, and for purposes of that theory relatives simply have to end up together. It doesn’t matter whether they are together because of shared habitat preference or because of true relative-recognition. For our present purpose, however, it is worth noting that, if further experiments confirm the rule of ‘incidental habitat selection’ for the tadpoles, this would rule out the ‘armpit’ theory but would not rule out the green-beard theory.

  Sherman (1979) invokes the idea of genetic favouritism in an ingenious theory about chromosome numbers in social insects. He presents evidence that eusocial insects tend to have higher chromosome numbers than their nearest non-social phylogenetic relatives. Seger (1980) independently discovered the effect, and has his own theory to account for it. The evidence for the effect is somewhat equivocal and would perhaps benefit from a demanding analysis using the statistical methods developed by modern students of the comparative method (e.g. Harvey & Mace 1982). But what I am concerned with here is not the truth of the effect itself but Sherman’s theory to account for it. He correctly notes that high chromosome numbers tend to reduce the variance in fraction of shared genes among siblings. To take an extreme case, if a species has only one chromosome pair with no crossing-over, any pair of full siblings will share (identical by descent) either all, none, or half of their genes with a mean of 50 per cent. If there are hundreds of chromosomes, on the other hand, the number of genes shared (identical by descent) among siblings will be narrowly distributed about the same mean of 50 per cent. Crossing-over complicates the issue, but it remains true that a high chromosome number in a species tends to go with low genetic variance among siblings in that species.

  It follows that if social insect workers wished to discriminate in favour of those of their siblings with whom they happened to share the most genes, it would be easier for them to do so if the species’s chromosome number were low than if it were high. Such preferential discrimination by workers would be detrimental to the fitness of the queen, who would ‘prefer’ a more even-handed treatment of her offspring. Sherman therefore suggests, in effect, that the high chromosome count in eusocial insects is an adaptation to cause “offsprings’ reproductive interests to more closely coincide with those of their mother”. We should not forget, by the way, that workers will not be unanimous. Each worker might show favouritism towards younger siblings that resemble her, but other workers will tend to resist her favouritism, for the same reason as the queen will resist it. The workers cannot be treated as a monolithic party in opposition to the queen, in the same way as Trivers and Hare (1976) were able to treat them in their theory of conflict over the sex ratio.

  Sherman very fairly lists three weaknesses of his hypothesis, but there are two more serious problems with it, Firstly, unless we are careful to qualify it further, the hypothesis appears to come dangerously close to a fallacy which I have dubbed ‘Misunderstanding Number 11’ (Dawkins 1979a) or the ‘Ace of Spades Fallacy’ (Chapter 10). Sherman assumes that the extent of cooperation among conspecifics is related to the ‘mean fraction of alleles they share’ (my emphasis) whereas he should think in terms of the probability that a gene ‘for’ cooperation is shared (see also Partridge & Nunney 1977). On the latter assumption his hypothesis, as at present expressed, would not work (Seger 1980). Sherman could rescue his hypothesis from this particular criticism by invoking the ‘armpit self-recognition’ effect. I shall not spell the argument out in detail, because I suspect that Sherman accepts it. (The essential point is that the armpit effect can use weak linkage within a family, while the green-beard effect requires pleiotropy or linkage disequilibrium. If workers inspect themselves, and show favouritism to those of their reproductive siblings that share features that they perceive themselves to possess, ordinary linkage effects become all-important and Sherman’s hypothesis could escape the ‘Ace of Spades Fallacy’. It would also, incidentally, escape the first of Sherman’s own objections, namely that the hypothesis ‘depends on the existence of alleles that enable their bearers to recognize their alleles’. ‘Such recognition alleles have never been discovered …’, and, by implication, are pretty implausible. Sherman could make things easier for himself by tying his hypothesis to the armpit rather than the green-beard effect.)

  My second difficulty with Sherman’s hypothesis was brought to my attention by J. Maynard Smith (personal communication). Taking the ‘armpit’ version of the theory, it is indeed conceivable that workers might be selected to inspect themselves, and show favouritism towards those of their reproductive siblings that share their own individual characteristics. It is also true that queens would then be selected to suppress this favouritism if they could, for instance by pheromonal manipulation. But, in order to be selected, any such move on the part of a queen would have to have an effect as soon as it arose as a mutation. Would this be true of a mutation that increased a queen’s chromosome number? No it would not. An increase in chromosome number would change the selection pressures bearing on worker favouritism, and many generations later it might produce an evolutionary change which would be to the advantage of queens in general. But this would not help the original mutant queen, whose workers would follow their own genetic programming and would be oblivious to changes in selection pressures. Changes in selection pressures exert their effects over the longer time-scale of generations. A queen cannot be expected to initiate a program of artificial selection for the long-term benefit of future queens! The hypothesis may be rescued from this objection by suggesting that high chromosome numbers are not an adaptation to facilitate queen manipulation of workers,
but rather are a preadaptation. Those groups that happened, for other reasons, to have high chromosome numbers, were the most likely to evolve eusociality. Sherman mentions this version of his hypothesis, but sees no reason to favour it over the more positive, maternal-manipulation version. In conclusion, Sherman’s hypothesis can be made theoretically sound, if it is phrased in terms of preadaptation rather than adaptation, and in terms of the armpit effect rather than the green-beard effect.

  The green-beard effect may be implausible, but it is instructive. The student of kin selection who first understands the hypothetical green-beard effect, and then approaches kin selection theory in terms of its similarities to and differences from ‘green-beard theory’, is unlikely to fall prey to the many tempting opportunities for error that kin selection theory offers (Dawkins 1979a). Mastery of the green-beard model will convince him that altruism towards kin is not an end in itself, something that animals are mysteriously expected to practise in accordance with some clever mathematics that field workers don’t understand. Rather, kinship provides just one way in which genes can behave as if they recognized and favoured copies of themselves in other individuals. Hamilton himself is emphatic on this point: ‘… kinship should be considered just one way of getting positive regression of genotype in the recipient, and … it is this positive regression that is vitally necessary for altruism. Thus the inclusive fitness concept is more general than “kin selection”’ (Hamilton 1975a, p. 140–141).