The Extended Phenotype
An extended phenotypic character is the product of the interaction of many genes whose influence impinges from both inside and outside the organism. The interaction is not necessarily harmonious—but then nor are gene interactions within bodies necessarily harmonious, as we saw in Chapter 8. The genes whose influences converge on a particular phenotypic character are a ‘physiological team’ only in a special and subtle sense, and this is true of the conventional within-body interactions to which Mayr refers, as well as of extended interactions.
I have previously tried to convey that special sense with the metaphor of a rowing crew (Dawkins 1976a, pp. 91–92), and with the metaphor of cooperation between myopic and normal-sighted people (Dawkins 1980, pp. 22–24). The principle might also be labelled the Jack Sprat principle. Two individuals with complementary appetites, say for fat and lean, or with complementary skills, say in growing wheat and milling it, form naturally harmonious partnerships, and it is possible to regard a partnership as a higher-order unit. The interesting question is how such harmonious units come about. I want to make a general distinction between two models of selective processes, both of which could, in theory, lead to harmonious cooperation and complementarity.
The first model invokes selection at the level of the higher-order units: in a metapopulation of higher-order units, harmonious units are favoured against disharmonious units. It was a version of this first model that I suggested was implicit in the Gaia hypothesis—selection among planets in that case. Coming down to earth, the first model might suggest that groups of animals whose members complement one anothers’ skills, say groups containing both farmers and millers, survive better than groups of farmers alone, or groups of millers alone. The second model is the one that I find more plausible. It does not need to postulate a metapopulation of groups. It is related to what population geneticists call frequency-dependent selection. Selection goes on at the lower level, the level of the component parts of a harmonious complex. Components within a population are favoured by selection if they happen to interact harmoniously with the other components that happen to be frequent in the population. In a population dominated by millers, individual farmers prosper, while in a population dominated by farmers it pays to be a miller.
Both kinds of model lead to a result which Mayr would call harmonious and cooperative. But I am afraid that the contemplation of harmony too often leads biologists to think automatically in terms of the first of the two models, and to forget the plausibility of the second. This is true of genes within a body just as it is true of farmers and millers in a community. The genotype may be a ‘physiological team’, but we do not have to believe that that team was necessarily selected as a harmonious unit in comparison with less harmonious rival units. Rather, each gene was selected because it prospered in its environment, and its environment necessarily included the other genes which were simultaneously prospering in the gene-pool. Genes with complementary ‘skills’ prosper in each others’ presence.
What does complementariness mean for genes? Two genes may be said to be complementary if the survival of each, relative to its alleles, is enhanced when the other is abundant in the population. The most obvious reason for such mutual assistance stems from the two genes performing a mutually complementary function within individual bodies that they happen to share. The synthesis of chemical substances of biological importance often depends upon a chain of steps in a biochemical pathway, each one mediated by a particular enzyme. The usefulness of any one of these enzymes is conditional upon the presence of the other enzymes in the chain. A gene-pool which is rich in genes for all enzymes in a given chain except one may set up a selection pressure in favour of the gene for the missing link in the chain. If there are alternative pathways to the same biochemical end product, selection may favour either pathway (but not both) depending upon initial conditions. Rather than regard the alternative pathways as units between which selection chooses (Model 1), it is better to think as follows (Model 2): selection will favour a gene that makes a given enzyme, to the extent that genes for making the other enzymes in its pathway are already abundant in the gene-pool.
But we do not have to stay at a biochemical level. Imagine a species of moth with stripes on the wings which resemble grooves in tree bark. Some individuals have transverse stripes while others, in a different area, have longitudinal stripes, the difference being determined at a single genetic locus. Clearly a moth will be well camouflaged only if it points itself in the right direction when sitting on tree bark (Sargent 1969b). Suppose some moths sit vertically and others horizontally, the behavioural difference being controlled at a second locus. An observer finds that, fortunately, all the moths in one area have longitudinal stripes and sit vertically, while in another area all the moths have transverse stripes and sit horizontally. We might say, then, that in both areas there is ‘harmonious cooperation’ between the genes for stripe orientation and the genes determining the orientation of sitting. How has this harmony come about?
Again we invoke our two models. Model 1 says that disharmonious gene combinations—transverse stripes with vertical resting behaviour, or longitudinal stripes with horizontal resting behaviour—died out, leaving only harmonious gene combinations. Model 1 invokes selection among combinations of genes. Model 2, on the other hand, invokes selection at the lower level of the gene. If, for whatever reason, the gene-pool in a given area happens to be already dominated by genes for transverse stripes, this will automatically set up a selection pressure at the behavioural locus in favour of genes for sitting horizontally. This in turn will set up a selection pressure to increase the predominance of transverse stripe genes at the striping locus which will, in turn, reinforce the selection in favour of sitting horizontally. The population will therefore rapidly converge on the evolutionarily stable combination, transverse stripes/sit horizontally. Conversely, a different set of starting conditions would lead the population to converge on the other evolutionarily stable state, longitudinal stripes/sit vertically. Any given combination of starting frequencies at the two loci will converge, after selection, on one or other of the two stable states.
Model 1 is applicable only if the pairs or sets of cooperating genes are especially likely to find themselves together in bodies, for instance if they are closely linked into a ‘supergene’ on one chromosome. They might indeed be so linked (Ford 1975), but Model 2 is particularly interesting because it enables us to visualize the evolution of harmonious gene complexes without such linkage. In Model 2 the cooperating genes may be on different chromosomes, and frequency-dependent selection will still lead to populations being dominated by genes that interact harmoniously with the other genes in the population, as a result of evolution to one or another evolutionarily stable state (Lawlor & Maynard Smith 1976). In principle the same kind of reasoning is applicable to sets of three loci (suppose stripes on the hindwing were controlled at a different locus from stripes on the forewing), four loci … n loci. If we try to model the interactions in detail the mathematics become difficult, but that does not matter for the point I want to make. All I am saying is that there are two general ways in which harmonious cooperation can come about. One way is for harmonious complexes to be favoured by selection over dis-harmonious complexes. The other is for the separate parts of complexes to be favoured in the presence, in the population, of other parts with which they happen to harmonize.
Having used Model 2 for the kind of within-body gene harmony that Mayr had in mind, we now generalize it to between-body, ‘extended’, gene interactions. We are going to be talking about genetic interaction at a distance, rather than the phenotypic action at a distance which was the subject of the earlier part of this chapter. This is easy to do, because frequency-dependent selection has classically been applied to between-body interactions, from Fisher’s (1930a) theory of the sex ratio on. Why do populations have a balanced sex ratio? Model 1 would suggest that it is because populations with an unbalanced sex ratio have gone extinct. Fisher’s own hypothesis
is, of course, a version of Model 2. If a population happens to have an unbalanced sex ratio, selection within that population favours genes that tend to restore the balance. There is no need to postulate a meta-population of populations, as in the case of Model 1.
Other examples of frequency-dependent advantage are well known to geneticists (e.g. Clarke 1979), and I have previously discussed their relevance to the controversy over ‘harmonious cooperation’ (Dawkins 1980, pp. 22–24). The point I want to stress here is that, from the point of view of each replicating entity, its relationships of harmony, cooperation and complementariness within genomes are not, in principle, different from relationships between genes in different genomes. The gene for sitting vertically on tree-trunks is favoured in a gene-pool which happens to be rich in genes for longitudinal stripes, and vice versa. Here, as in the biochemical example of the chain of enzymes, the cooperation takes place within bodies: the significance of the fact that the gene-pool is rich in genes for longitudinal stripes is that any given gene at the locus determining sitting behaviour is therefore statistically likely to be in a longitudinally striped body. I suggest that we should think primarily of genes as being selected against the background of the other genes that happen to be frequent in the gene-pool, and only secondarily make a distinction between whether the salient between-gene interactions happen to occur within bodies or between them.
Wickler (1968), in his fascinating review of animal mimicry, points out that individuals sometimes appear to cooperate in achieving mimetic resemblance. He recounts an observation by Koenig of what appeared to be a sea anemone in an aquarium tank. The following day there were two anemones, each half the size of the original one, and the day after that the original large anemone had apparently reconstituted itself. The impossibility of this finally led Koenig to investigate in detail, and he discovered that the ‘anemone’ was in fact a fake, put together by numerous cooperating annelid worms. Each worm represented one tentacle, and they grouped themselves into a circle in the sand. Fish seemed to be fooled by the deception just as Koenig originally was, for they gave the fake anemone the same kind of wide berth they would a real one. Each individual worm presumably gained protection from fish predators, by joining in the cooperative mimicry ring. I suggest that it is not helpful to speak of groups of worms that form rings as being selected over groups that do not. Rather, ring-joining individuals are favoured in populations of ring-joiners.
In various insect species, each individual mimics one flower of a many-flowered inflorescence, and therefore a cooperating crowd of them is needed before a whole inflorescence can be convincingly mimicked. ‘In East Africa it is possible to find a particular plant with extremely beautiful inflorescences … the individual flowers are about half a centimetre in height, look rather like broom flowers, and are arranged around a vertical stem like the flowers of the lupin. Experienced botanists have taken this plant for Tinnaea or Sesamopteris and found themselves suddenly holding a bare stem after plucking the “flower”. The flower had not fallen off—it had flown away! The “flower” consists of cicadas, either Ityraea gregorii or Oyarina nigritarsus’ (Wickler 1968, p. 61).
In order to develop my argument I need to make certain detailed assumptions. Since the details of the selection pressures bearing on these particular cicada species are not known, it will be safest if I invent a hypothetical cicada which basically practises the same group-mimicry trick as Ityraea and Oyarina. I assume that my species occurs in two colour morphs, pink and blue, and that the two morphs mimic two different colour varieties of lupin. Pink and blue lupins are assumed to be equally abundant over the whole range of the cicada species, but in any one local area all the cicadas are either pink or blue. ‘Cooperation’ occurs, in that individuals cluster together near the tips of plant stems and together resemble lupin inflorescences. It is ‘harmonious’ in that mixed colour clusters do not occur: I assume that a mixed colour cluster is especially likely to be spotted by predators as a fake, since real lupins do not have two-tone inflorescences.
Here is how the harmony might have come about through Model 2’s frequency-dependent selection. In any given area, historical accident determined that there was an initial majority in favour of one colour type or the other. In an area that happened to be dominated by pink cicadas, blue ones were penalized. In an area that happened to be dominated by blue cicadas, pink ones were penalized. In both cases, simply being in the minority was disfavoured, because a member of the minority type was, by the laws of chance, more likely than a member of the majority type to find itself participating in a mixed cluster. At the gene level we may say that pink genes are favoured in a gene-pool dominated by pink genes, and blue genes are favoured in a gene-pool dominated by blue genes.
We now invent another insect, say a caterpillar, which is large enough to mimic a whole lupin inflorescence, instead of a single flower. Each segment of the caterpillar mimics a different flower of the inflorescence. The colour of each segment is controlled at a different locus, the alternatives being pink and blue. A caterpillar that is all blue or all pink is more successful than one that is a mixture of colours because, once again, predators have learned that mixed colour lupins do not occur. There is no theoretical reason why two-tone caterpillars should not occur, but suppose that, as a result of selection, they do not: in any one area the local caterpillars are either all pink or all blue. We have ‘harmonious cooperation’ again.
How might the harmonious cooperation come about? By definition, Model 1 would be applicable only if the genes responsible for the coloration of the different segments were linked tightly in a supergene. Multi-coloured supergenes would be penalized at the expense of pure pink and pure blue supergenes. In this hypothetical species, however, the relevant genes are widely spread on different chromosomes, and we have to apply Model 2. In any given local area, once one colour starts to predominate at a majority of loci, selection works to increase the frequency of that colour at all loci. In a particular area, if all the loci save one are dominated by pink genes, the odd locus which is dominated by blue genes will soon be brought into line by selection. As in the case of the hypothetical cicadas, historical accidents in different local areas automatically set up selection pressures in favour of one or another of two evolutionarily stable states.
The point of this thought experiment is that Model 2 is equally applicable both between and within individuals. In both the caterpillar and the cicada case, pink genes are favoured in gene-pools already dominated by pink genes, and blue genes are favoured in gene-pools already dominated by blue genes. In the caterpillars the reason is that each gene benefits if it shares a body with other genes producing the same colour as itself. In the cicadas the reason is that each gene benefits if the body it is in meets another body bearing a gene producing the same colour as itself. In the caterpillar example, the cooperating genes occupy different loci in the same individual. In the cicada example the cooperating genes occupy the same locus in different individuals. My purpose is to close the conceptual gap between these two kinds of gene interaction, by showing that genetic interaction at a distance is not, in principle, different from genetic interaction within one body.
To resume my series of quotations from Mayr:
The result of the coadapting selection is a harmoniously integrated gene complex. The coaction of the genes may occur at many levels, that of the chromosome, nucleus, cell, tissue, organ, and whole organism.
The reader will by now have no difficulty in guessing how Mayr’s list is to be extended. Coaction among genes in different organisms is not fundamentally different from coaction among genes in the same organism. Each gene works in a world of phenotypic consequences of other genes. Some of those other genes will be members of the same genome. Others will be members of the same gene-pool operating through other bodies. Yet others may be members of different gene-pools, different species, different phyla.
The nature of the functional mechanisms of physiological interaction are [sic] only of
minor interest to the evolutionist, whose main concern is the viability of the ultimate product, the phenotype.
Mayr hits the nail on the head again, but his ‘phenotype’ is not the ultimate: it can be extended outside the individual body.
Many devices tend to maintain the status quo of gene pools, quantitatively and qualitatively. The lower limit of genetic diversity is determined by the frequent advantage of heterozygosity … The upper limit is determined by the fact that only those genes can be incorporated that are able to ‘coadapt’ harmoniously. No gene has a fixed selective value; the same gene may confer high fitness on one genetic background and be virtually lethal on another.
Excellent, but remember that ‘genetic background’ can include genes in other organisms as well as genes within the same organism.
The result of the close interdependence of all genes in a gene pool is tight cohesion. No gene frequency can be changed, nor any gene be added to the gene pool, without an effect on the genotype as a whole, and thus indirectly on the selective value of other genes.
Mayr himself has now subtly shifted to talking about a coadapted gene pool, rather than a coadapted individual genome. This is a great step in the right direction, but we must still take one further step. Mayr is here talking about interactions between all the genes in one gene-pool, regardless of the bodies they happen to be sitting in. The doctrine of the extended phenotype ultimately requires us to acknowledge the same kind of interactions among genes of different gene-pools, different phyla, different kingdoms.
Consider again the ways in which a pair of genes in the same gene-pool can interact, more specifically the ways in which the frequency of each in the gene-pool can affect the survival prospects of the other. The first way, and the one which I suspect Mayr had mainly in mind, is through sharing the same body. The survival prospects of gene A are influenced by the frequency in the population of gene B, because B’s frequency influences the probability that A will find itself sharing a body with B. The interaction between the loci determining moth stripe direction and sitting direction was an example of this. So was the hypothetical lupin-mimicking caterpillar. So is a pair of genes coding for enzymes that are necessary for successive stages in a particular pathway synthesizing a useful substance. Call this type of gene interaction ‘within-body’ interaction.