The Bactra Review: Occasional and eclectic book reviews by Cosma Shalizi   115

The Extended Phenotype

The Long Reach of the Gene

by Richard Dawkins

with a new afterword by Daniel Dennett

Oxford University Press, 1999

Not Your Parents' Web of Life

In the 19th century, Thomas Huxley was known as "Darwin's bull-dog" for his active and tenacious defense of Darwinism. Richard Dawkins fills a similar niche today, only with much more force and much less reserve about calling a spade a shovel --- he's Darwin's pit-bull, not his bull-dog. Books like The Blind Watchmaker and Climbing Mount Improbable do an immense amount of good in the world, and I've seen them turn the minds of many readers inside out. But they also lead to the reasonable suspicion that Dawkins doesn't have much to offer those who accept what are, after all, the elementary facts of life. By Dawkins's own reckoning, his most important book is this one, and he's right. It gives what is at once the most forceful and the most considered presentation of his distinctive ideas, and it contain a genuinely important and novel concept derived from them, namely that of the extended phenotype itself.

It is only fitting, then, that this epitome of Dawkins opens with one of the most impressive refutations of genetic determinism I have encountered, at least as powerful as any Stephen Jay Gould or Richard Lewontin has ever produced. Genes do not determine anatomy or physiology or behavior; genes encode the information to make proteins. Which genes will be expressed --- decoded into proteins --- depends on a very subtle biochemical process, the complexity and intricacy of which is hidden by the labels "signal transduction" and "genetic regulation". The idea that, once we know an organism's genotype --- have a complete inventory of its genes --- we could read off its traits is a baseless fantasy propagated by people who are at best idiots (science journalists, movie writers, newspaper editors, professional futurists), or at worst should know much better (science fiction novelists, some molecular biologists involved in the Human Genome Project).

What then is the sense of speaking of a gene for a trait, as competent evolutionists and geneticists do all the time? Essentially, it's a statistical claim about causation. A gene X is "for" a trait Y if and only if different versions of X --- different alleles of X --- lead to differences in the distribution of Y, in a particular context of other genes and environmental factors. We can de-relativize this either by taking that context as more or less fixed, or by saying that X is for Y if there is any context in which differences in X lead to differences in Y. Now, this statement does not commit us to the genetic determination of Y, at least not in any sane sense of "determination". In particular, it is not a claim that:

A gene-centered view of evolution does not depend on any of these things. All that the view needs are the facts that (1) genes are capable of statistical influence on traits; (2) some of those traits influence reproductive success; (3) genes pass mostly intact through reproduction --- they are high-fidelity replicators; (4) other things --- genomes, whole organisms, groups of organisms, species --- do not satisfy these constraints. Now, (1)--(3) are certainly true, and not in dispute among those with their heads plugged in. It is not hard --- now --- to see that the conjunction of these premises will lead to an increase in the relative frequency of replicators which --- in the context of other replicators and environmental contingencies --- produce traits which have tended to aid reproductive success in the past. That is, there will be selection for genes which act as if they were interested in increasing their representation in future generations, as if they pursued selfish genetic interests, as if genes wanted to make more genes in their image, and produced "vehicles" which serve that end. I will henceforth speak about genes making organisms do things which serve the genes' interests, with the understanding that these metaphors can all be replaced, at the price of tedium, with non-metaphorical statements about the statistical effects on representation in future generations of allelic differences in a given genetic-environmental context (a disclaimer which itself illustrates the use of the metaphors).

The real controversy is in the claim (4), that organisms, groups and species are not replicators, or at least not evolvable replicators. Groups can produce more groups, and species can produce more species, just as genes can produce more genes; does a process analogous to gene natural selection act on these levels? This is an empirical issue, not an a priori one. Do these putative replicators have a strong influence on traits which themselves have a strong influence on reproductive success? That is, are these other replicators under strong selection pressure? Is the copying fidelity of these replicators high? The stronger the selection pressure, the sloppier replication can be and still support natural selection. Is selection acting at these levels strong enough to overcome the effects of gene selection and random drift? It seems that groups, species, etc., are both under very weak selection pressure and very sloppy replicators. But, as I said, genes are under strong selection pressure and are high-fidelity replicators; selection on genes, therefore, rules.

This leaves open the question of whether, in some circumstances, selection might favor genes which, while deleterious to their carriers, benefit other members of their group, even without kin selection (on which see below). It's been shown --- for instance by John Pepper and Barbara Smuts --- that this is possible, but I don't think we know whether it's important in nature. This is sometimes also called "group selection," but I think that should be reserved for theories in which groups replicate.

One consequence of taking the gene's-eye-view is that we should expect the versions of genes which encounter each other often --- the most common alleles in the gene pool --- to be ones which are well-adapted to each other; indeed, co-adapted. That is, genes will be selected to have consequences favorable to their own reproduction in the presence of common alleles. (There's a clear route from here to evolutionary game theory and evolutionarily stable strategies.) This does not mean that they will be mutually supportive, however. Rabbit genes will tend to adapt to a context partially determined by which genes are common among foxes, and vice versa: hence evolutionary arms races, on which Dawkins is (as usual) particularly acute.

Genes are selfish; if one can prosper replicatively at the expense of others, it will do so. Normally, organisms have fairly elaborate machinery which ensure that either all their genes get reproduced or none of them do; this puts all their genes in the same boat, and encourages their cooperation. It also, however, gives a serious advantage to any gene which can subvert that machinery in its own favor. (Variants which muck with the machinery to their detriment are, obviously, not long for this world.) There is considerable evidence (reviewed here) that such extra-selfish genes are actually fairly common, and somewhat less evidence that they do not completely dominate natural populations because all genes not linked to the "outlaws" will be selected to modify and inhibit their effects --- the "parliament of genes" effect.

Not only can genes within a single body compete, but genes in different bodies can collaborate. The most familiar example of this is kin selection, the phenomenon encapsulated in Haldane's joke that he would "lay down his life for two brothers or eight cousins." That is, a gene might find it in its interest to make the body it is in do things which depress its reproductive prospects, if by doing so it raises those of other bodies which contain the gene. In fact, if the benefits to the other bodies are sufficiently large in relation to the costs, it makes sense for the gene to do this even if the other bodies only probably contain the gene.

Mention of kin selection brings up the problem of "fitness". Many readers will be surprised to learn that Dawkins thinks it is a not particularly useful term, and one which evolutionist would probably be better off discarding. In his ch. 10, "An Agony in Five Fits," he distinguishes no less than five senses in which the term is used, only one of them (the fitness of a genotype in population genetics) reasonably useful and operational. This is probably the most technically involved chapter of the book, but also one of the most important in driving home the point that selection really does act on genes, the true replicators, and nothing else. This concludes the critical and as it were dissecting portion of the book; having convinced his readers that genes are what matters, he now (like the good reductionist that he is) proceeds to show how they interact with each other to give us the phenomena we know and love, or at least suffer through. Some of this work has actually already been done, in the way of looking at arms-races, the co-adaptation of genes in a common gene pool, and so forth. But the consequences of having genes interact are, literally, much more far-ranging.

The phenotype of an organism is the collection of all its "manifested attributes" --- a clear enough notion, if one difficult to make metaphysically precise. The phenotype is the joint product of the genotype and the environment. The original and important point Dawkins makes under the label of "the extended phenotype" is that there is no good reason to suppose that the phenotype stops at the skin (or bark or what-not), and that there are many aspects of life which we can account for straightforwardly if only we suppose that it does not. Dawkins starts modestly and plausibly enough with the effects an organism has on its inanimate environment. Many animals make artifacts --- shells, burrows, stores, and what-not --- which are subject to genetic variation, and there seems to be no principled reason not to regard the relevant genes as being "for" those artifacts, just as other genes are for bodily traits. If variation in the artifacts in turn effects reproductive success, then those genes will, in fact, be exposed to natural selection in the ordinary way. (So far as I can recall, all of Dawkins's examples are of animal behavior, but I don't see why we shouldn't expect plants, bacteria and the like to have extended phenotypes as well. A tree which by increasing shade on the forest floor inhibits the growth of rival plants would be a simple example.)

What is even more interesting is that Dawkins shows how the same mechanisms can work even in the case of artifacts which are the joint work of many animals.

Beaver dams and termite mounds are collectively built by the behavioural efforts of more than one individual. A genetic mutation in one individual beaver could show itself in phenotypic change in the shared artefact. If the phenotypic change in the artefact had an influence on the success of replication of the new gene, natural selection would act, positively or negatively, to change the probability of similar artefacts existing in the future. The gene's extended phenotypic effect, say an increase in the height of the dam, affects its chances of survival precisely in the same sense as in the case of a gene with a normal phenotypic effect, such as an increase in the length of the tail. The fact that the dam is the shared product of the building behaviour of several beavers does not alter the principle: genes that tend to make beavers build high dams will themselves, on average, tend to reap the benefits (or costs) of high dams, even though every dam may be jointly built by several beavers. If two beavers working on the same dam have different genes for dam height, the resulting extended phenotype will reflect the interaction between the genes, in the same way as bodies reflect gene interactions. There could be extended genetic analogues of epistasis, of modifier genes, even of dominance and recessiveness. [p. 209]
In fact, Dawkins even shows that genes for a common extended-phenotypic trait can be spread over several species, even over several kingdoms, without messing up the mechanism of their evolution.

The next step is to go from phenotypes which are produced by several bodies to phenotypes in other bodies. The most obvious case is that of parasites altering the behavior of their hosts. This gives Dawkins a multitude of bizarre examples of body-snatching, in which he plainly revels; but the subject is also of great importance to medicine, where we want to know, before trying to suppress a symptom, whether it is the host's way of combating the parasite, or on the contrary something the parasite makes the host do to the parasite's advantage. Having been lulled by the parasites, the reader is prepared to see extended-phenotypic manipulations all over the place: between parents and offspring, for instance (in either direction), or between siblings (in eusocial insects, for instance); the field of animal communication as a whole begins to look like an extended exercise in indirect mind-control. Dawkins is at length led to the following thesis: "An animal's behaviour tends to maximize the survival of the genes 'for' that behaviour, whether or not those genes happen to be in the body of the particular animal performing it" (p. 233). Do not be misled by fact that this sentence is in the indicative; it is really a recommendation to look for genes which benefit from behaviors, whether they happen to be in the same body or not, and not a statement that, as a matter of fact, there are always such genes (cf. ch. 3, "Constraints on Perfection").

Ultimately, Dawkins presents a vision of the organic world and its appurtenances as overlapping fields of power exerted by replicators over each other and over the vehicles which they construct to carry themselves into future generations. Replicators, moreover, are fully capable of (as Dawkins puts it) "action at a distance," and potentially immense distances in both space and time, by quite subtle routes. It is a tough-minded vision, deeply at odds with the view that nature is naturally cooperative and harmonious, except where we've mucked it up --- the view that Dawkins, with his usual accurate scorn, calls "the BBC Theorem". The web of genetic control and interdependence is as vast, subtle and far-reaching as that invoked by the BBC Theorem, but immeasurably less nice, allowing both cooperation and competition, as the interests of the genes dictate. (I shall refrain from pointing out the similarities between this vision and that of Foucault, since that would merely cause indignation all around.)

The last chapter is devoted to the question of why, given this vision of replicative action-at-a-distance, life should come bundled into discrete organisms at all. Passing on the question of why we have cells, Dawkins turns to the problem of why multicellular organisms should exist, and why they should all go through a stage where they consist of a single cell which then multiplies and differentiates. He presents an ingenious argument, derived from one put forth by by the developmental biologist John Tyler Bonner, as to why this "bottle-neck" facilitates the evolution of complex adaptations, in ways that simple growth would not. This doesn't guarantee such bottlenecks will evolve, but shifts the problem to the more tractable one of why they do so. (The latest views on this have been summarized by Maynard Smith and Szathmáry, as constant readers will recall.)

This is a technical and controversial work addressed to Dawkins's fellow biologists; it is also enviably, relentlessly clear, both in its prose and its logic. This is writing like a spear: a hard, sharp point, everything needed to drive the point home, and nothing else. Inevitably it needs more background than a book like The Blind Watchmaker, but it can be followed by those whose knowledge of evolutionary biology is entirely at the level of popular works. I strongly urge everyone with that preparation to read this book, not only to encounter an important, coherent view of evolution, but simply to see what can be done with words and ideas in the hands of a master.


Disclaimer: I didn't ask for a review copy of this book, but the publishers sent me one anyway. Needless to say, I have no stake in its success.
x + 313 pp., glossary, bibliography, "further reading" (= selected relevant works which have appeared since 1982), author and subject indices
Evolution / Genetics
Currently in print as a paperback, US$16.95, ISBN 0-19-288051-9 [Buy from Powell's]. Apart from the "further reading" and the afterword, this edition is identical to the edition of 1989, which in turn differed from that of 1982 only in the subtitle, and in the addition of two paragraphs to the preface.
25 February 2000
Thanks to John Pepper for a pre-print copy of his paper and helpful discussions, and to Danny Yee for pointing out an obscurity.