The Extended Phenotype

by Richard Dawkins

  Once again, if this is to be regarded as a parasitic adaptation, and indeed it is widely so regarded (Wickler 1968; Holmes & Bethel 1972), we are forced to postulate the sometime existence of genes in the parasite gene-pool that influenced the behaviour of hosts, since all Darwinian adaptations evolved by the selection of genes. By definition such genes were genes ‘for’ snail behaviour, and the snail behaviour has to be regarded as part of the phenotypic expression of fluke genes.

  The genes in one organism’s cells, then, can have extended phenotypic influence on the living body of another organism; in this case a parasite’s genes find phenotypic expression in the behaviour of its host. The parasitology literature is full of interesting examples which are now usually interpreted as adaptive manipulation of hosts by parasites (e.g. Holmes & Bethel 1972; Love 1980). It has not, to be sure, always been fashionable among parasitologists to make such interpretations explicit. For instance, an important review of parasitic castration in Crustacea (Reinhard 1956) is packed with detailed information and speculation on the precise physiological routes by which parasites castrate their hosts, but is almost devoid of discussion on why they might have been selected to do so, or whether, instead, castration is simply a fortuitous byproduct of parasitization. It may be an interesting indication of changing scientific fashions that a more recent review (Baudoin 1975) extensively considers the functional significance of parasitic castration from the individual parasite’s point of view. Baudoin concludes: ‘The main theses of this paper are (1) that parasitic castration can be viewed as a parasite’s adaptation and (2) that advantages derived from this adaptation are a result of reduction in host reproductive effort; which in turn gives rise to increased host survivorship, increased host growth and/or increased energy available to the parasite thereby increasing the parasite’s Darwinian fitness.’ This is, of course, very much the argument I have just followed in discussing parasite-induced thickening of snail shells. Again, a belief that parasitic castration is a parasite adaptation logically implies that there must be, or at least must have been, parasite genes ‘for’ changes in host physiology. The symptoms of parasitic castration, sex-change, increased size, or whatever they may be, are properly regarded as extended phenotypic manifestations of parasite genes.

  The alternative to Baudoin’s interpretation is that changes in host physiology and behaviour are not parasite adaptations, but simply dull pathological byproducts of infection. Consider the parasitic barnacle (though in its adult stage it looks more like a fungus) Sacculina. It might be said that Sacculina does not directly benefit from the castration of its crab host, but simply sucks nutriment from all through the host’s body, and as a side-effect of its devouring the tissues of the gonads the crab shows the symptoms of castration. In support of the adaptation hypothesis, however, Baudoin points to those cases in which parasites achieve the castration by synthesis of host hormones, surely a specific adaptation rather than a boring byproduct. Even in those cases where the castration is initially caused as a byproduct of the devouring of gonadal tissue, my suspicion is that selection would subsequently act on parasites to modify the details of their physiological effects on hosts, modify them in a way favourable to the parasite’s well-being. Sacculina presumably has some options over which parts of the crab’s body its root system invades first. Natural selection is surely likely to favour genes in Sacculina that cause it to invade gonadal tissue before invading vital organs on which the crab’s survival depends. Applying this kind of argument at a more detailed level, since gonadal destruction has multiple and complex effects on crab physiology, anatomy and behaviour, it is entirely reasonable to guess that selection would act on parasites to fine-tune their castration technique so as to increase their benefit from the initially fortuitous consequences of castration. I believe many modern parasitologists would agree with this feeling (P. O. Lawrence, personal communication). All that I am adding, then, is the logical point that the common belief that parasitic castration is an adaptation implies that the modified host phenotype is part of the extended phenotype of parasite genes.

  Parasites often stunt the growth of their hosts, and it is easy to see this as a boring byproduct of infection. More interest therefore attaches to those rarer cases in which parasites enhance host growth, and I have already mentioned the case of the thickened snail shells. Cheng (1973, p. 22) begins his account of such cases with a revealing sentence: ‘Although one generally considers parasites to be detrimental to their hosts and cause the loss of energy and poor health, instances are known where the occurrence of parasites actually induces enhanced growth of the host.’ But Cheng is here sounding like a medical doctor rather than like a Darwinian biologist. If ‘detriment’ is defined in terms of reproductive success rather than survival and ‘health’, it is probable that growth-enhancement is indeed detrimental to the host, for the reasons given in my discussion of snail shells. Natural selection has presumably favoured an optimal host size, and if a parasite causes the host to deviate from this size in either direction it is probably harming the host’s reproductive success, even if it is at the same time promoting host survival. All the examples of growth-enhancement given by Cheng can be easily understood as parasite-induced switches of resources away from investment in host reproduction, which is of no interest to the parasite, to the growth and survival of the host’s own body, which is of great interest to the parasite (again we must beware of the group-selectionist plea that the existence of a new generation of hosts is of importance to the parasite species).

  Mice infected by larvae of the tapeworm Spirometra mansanoides grow faster than uninfected mice. It has been shown that the tapeworms achieve this by secreting a substance resembling a mouse growth hormone. More dramatically, beetle larvae of the genus Tribolium, when infected by the sporozoan Nosema, usually fail to metamorphose into adults. Instead they continue to grow through as many as six extra larval moults, ending up as giant larvae weighing more than twice as much as non-parasitized controls. Evidence suggests that this major shift in beetle priorities, from reproduction to individual growth, is due to the synthesis of juvenile hormone, or its close analogue, by the protozoan parasite. Again this is of interest because, as already suggested for the case of parasitic castration of crustacea, it makes the boring byproduct theory all but untenable. Juvenile hormones are special molecules ordinarily synthesized by insects, not protozoa. The synthesis of an insect hormone by a protozoan parasite has to be regarded as a specific and rather elaborate adaptation. The evolution of the capacity to synthesize juvenile hormone by Nosema must have come about through the selection of genes in the Nosema gene-pool. The phenotypic effect of these genes, which led to their survival in the Nosema gene-pool, was an extended phenotypic effect, an effect which manifested itself in beetle bodies.

  Once again the problem of individual benefit versus group benefit arises, and in an acute form. A protozoan is so small in comparison with a beetle larva that a single protozoan, on its own, could not muster a sufficient dose of hormone to affect the beetle. Hormone manufacture must be a group effort by large numbers of individual protozoa. It benefits all individual parasites in the beetle, but it must also cost each individual something to add his tiny contribution to the group chemical effort. If the individual protozoa were genetically heterogeneous, consider what would happen. Assume a majority of protozoa cooperating in synthesizing hormone. An individual with a rare gene that made him opt out of the group effort would save himself the cost of synthesis. Such a saving would be of immediate benefit to him, and to the selfish gene that made him opt out. The loss of his contribution to the group synthesis would hurt his rivals just as much as it hurt him. In any case the loss to the group’s productivity would be very small, though it would represent a major saving to him. Therefore, except under special conditions, taking part in a cooperative group synthesis together with genetic rivals is not an evolutionarily stable strategy. We must therefore predict that all the Nosema in a given beetle will be found to be
close relatives, probably an identical clone. I don’t know of any direct evidence here, but the expectation is in line with the typical sporozoan life cycle.

  Baudoin properly emphasizes the analogous point in connection with parasitic castration. He has a section entitled ‘Kinship of castrators in the same individual host’, in which he says ‘Parasitic castration is almost invariably produced either by single parasites or by their immediate offspring … parasitic castration is usually produced either by single genotypes or by very closely related genotypes. Metacercarial infections in snails are exceptions … In these cases, however, parasitic castration may be incidental.’ Baudoin is fully aware of the significance of these facts: ‘… the genetic relationship of castrators within individual hosts is such that natural selection at the level of individual genotypes can account for the observed effects’.

  Many fascinating examples of parasites manipulating the behaviour of their hosts can be given. For nematomorph larvae who need to break out of their insect hosts and get into water where they live as adults, ‘… a major difficulty in the parasite’s life is the return to water. It is, therefore, of particular interest that the parasite appears to affect the behaviour of its host, and “encourages” it to return to water. The mechanism by which this is achieved is obscure, but there are sufficient isolated reports to certify that the parasite does influence its host, and often suicidally for the host … One of the more dramatic reports describes an infected bee flying over a pool and, when about six feet over it, diving straight into the water. Immediately on impact the gordian worm burst out and swam into the water, the maimed bee being left to die’ (Croll 1966).

  Parasites that have a life cycle involving an intermediate host, from which they have to move to a definitive host, often manipulate the behaviour of the intermediate host to make it more likely to be eaten by the definitive host. We have already seen an example of this, the case of Leucochloridium in snail tentacles. Holmes and Bethel (1972) have reviewed many examples, and they have themselves provided us with one of the most thoroughly researched cases (Bethel & Holmes 1973). They studied two species of acanthocephalan worms, Polymorphus paradoxus and P. marilis. Both use a freshwater ‘shrimp’ (really an amphipod), Gammarus lacustris, as an intermediate host, and both use ducks as the definitive host. P. paradoxus, however, specializes in the mallard, which is a surface-dabbling duck, while P. marilis specializes in diving ducks. Ideally then, P. paradoxus might benefit by making its shrimps swim to the surface, where they are likely to be eaten by mallards, while P. marilis might benefit by making its shrimps avoid the surface.

  Uninfected Gammarus lacustris tend to avoid light, and stay close to the lake bottom. Bethel and Holmes noticed striking differences in the behaviour of shrimps infected with cystacanths of P. paradoxus. They stayed close to the surface, and clung tenaciously to surface plants and even to the hairs on the researchers’ legs. This surface-hugging behaviour would presumably make them vulnerable to predation by dabbling mallards, and also by muskrats which are an alternative definitive host of P. paradoxus. Bethel and Holmes believe that the habit of clinging to weeds makes infected shrimps particularly vulnerable to muskrats, who gather floating vegetation and take it home to feed off it.

  Laboratory tests confirmed that shrimps infected with cystacanths of P. paradoxus seek the lighted half of a tank, and also positively approach the source of light. This is the opposite of the behaviour shown by uninfected shrimps. Incidentally, it is not that the infected shrimps were generally sick and floated passively to the surface, as in the comparable case of Crowden and Broom’s (1980) fish. These shrimps fed actively, often leaving the surface layer to do so, but when they captured a morsel of food they promptly took it to the surface to eat it whereas a normal shimp would have taken it to the bottom. And when startled in mid water, instead of diving to the bottom as a normal shrimp would, they headed for the surface.

  Shrimps infected by cystacanths of the other species, P. marilis, however, do not hug the surface. In laboratory tests they admittedly sought the lighted half of an aquarium in preference to the dark half, but they did not orient positively towards the source of light: they distributed themselves randomly in the lighted half, rather than at the surface. When startled, they went to the bottom rather than to the surface. Bethel and Holmes suggest that the two species of parasite modify the behaviour of their intermediate host in different ways, calculated to make the shrimps more vulnerable to predation by their definitive hosts, surface-feeding and diving predators respectively.

  A later paper (Bethel & Holmes 1977) provides partial confirmation of this hypothesis. Captive mallards and muskrats in the laboratory took shrimps infected with P. paradoxus at a higher rate than they took uninfected shrimps. Shrimps infected with P. marilis, however, were not taken any more often than uninfected ones, either by mallards or by muskrats. It would obviously be desirable to do the reciprocal experiment with a diving predator, predicting that shrimps infected with cystacanths of P. marilis would, in this case, be relatively more vulnerable. This experiment does not appear to have been done.

  Let us provisionally accept Bethel and Holmes’s hypothesis, and rephrase it in the language of the extended phenotype. The altered behaviour of the shrimp is regarded as an adaptation on the part of the acanthocephalan parasite. If this has come about through natural selection, there must have been genetic variation ‘for’ shrimp behaviour in the worm gene-pool, otherwise there would have been nothing for natural selection to work on. We may, therefore, talk of worm genes having phenotypic expression in shrimp bodies, in just the same sense as we are accustomed to talking of human genes having phenotypic expression in human bodies.

  The case of the fluke (‘brainworm’) Dicrocoelium dendriticum is often quoted as another elegant example of a parasite manipulating an intermediate host to increase its chances of ending up in its definitive host (Wickler 1976; Love 1980). The definitive host is an ungulate such as a sheep, and the intermediate hosts are first a snail and then an ant. The normal life cycle calls for the ant to be accidentally eaten by the sheep. It seems that the fluke cercaria achieves this in a way analogous to that of the Leucochloridium mentioned above. By burrowing into the suboesophageal ganglion, the aptly named ‘brainworm’ changes the ant’s behaviour. Whereas an uninfected ant would normally retreat into its nest when it became cold, infected ants climb to the top of grass stems, clamp their jaws in the plant and remain immobile as if asleep. Here they are vulnerable to being eaten by the worm’s definitive host. The infected ant, like a normal ant, does retreat down the grass stem to avoid death from the midday heat—which would be bad for the parasite—but it returns to its aerial resting position in the cool of the afternoon (Love 1980). Wickler (1976) says that of the approximately fifty cercariae that infect a given ant, only one burrows into the brain and it dies in the process: ‘It sacrifices itself for the benefit of the other cercariae.’ Understandably, Wickler therefore predicts that the group of cercariae in an ant will be found to be a polyembryonic clone.

  An even more elaborate example is the case of crown gall, one of the few known plant cancers (Kerr 1978; Schell et al. 1979). Exceptionally for a cancer it is induced by a bacterium, Agrobacterium. These bacteria induce the cancer in the plant only if they themselves contain a Ti plasmid, a small ring of extrachromosomal DNA. The Ti plasmid may be regarded as an autonomous replicator (Chapter 9) although, like any other DNA replicator, it cannot succeed apart from the cellular machinery put together under the influence of other DNA replicators, in this case those of the host. Ti genes are transferred from bacterial to plant cells, and the infected plant cells are induced to multiply uncontrollably, which is why the condition is called a cancer. The Ti genes also cause the plant cells to synthesize large quantities of substances called opines, which the plants would not normally make, and which they cannot use. The interesting point is that, in an environment rich in opines, Ti-infected bacteria survive and reproduce much better than non-infected
bacteria. This is because the Ti plasmid provides the bacterium with a set of genes which enable the bacterium to use the opines as a source of energy and chemicals. The Ti plasmids could almost be regarded as practising artificial selection in favour of infected bacteria, hence in favour of copies of themselves. The opines also function as bacterial ‘aphrodisiacs’, as Kerr puts it, promoting bacterial conjugation and therefore plasmid transfer.

  Kerr (1978) concludes: ‘It is a very elegant example of biological evolution; it even demonstrates apparent altruism in bacterial genes … The DNA which is transferred from bacterium to plant has no future; it dies when the plant cell dies. However, by altering the plant cell to produce an opine, it ensures (a) preferential selection of the same DNA in bacterial cells and (b) transfer of that DNA to other bacterial cells. It demonstrates evolution at the level of genes, not organisms, which may only be gene carriers.’ (Such statements are, of course, music to my ears, but I hope Kerr will forgive my publicly wondering at the gratuitous caution of ‘may only be’ gene carriers. It is a bit like saying ‘The eyes may be the windows of the soul’ or ‘O, my Luve may be like a red red rose’. It may have been the work of an editorial hand!) Kerr continues: ‘In naturally-induced crown galls on many (but not all) hosts, very few bacteria survive in the gall … At first sight, it would appear that pathogenicity is conferring no biological advantage. It is only when one considers opine production by the host and its effect on bacteria living at the surface of the gall, that the strong selective advantage of genes for pathogenicity becomes clear.’

  Mayr (1963, pp. 196–197) discusses the phenomenon of plants making galls to house insects, in terms so favourable to my thesis that I can quote him verbatim almost without comment: