The Blind Watchmaker

by Richard Dawkins

  So long as the molecular clock is a fact — and it does seem to be true that each kind of molecule changes at roughly its own characteristic rate per million years — we can use it to date branch points in the evolutionary tree. And if it is really true that most evolutionary change, at the molecular level, is neutral, this is a wonderful gift for the taxonomist. It means that the problem of convergence may be swept away by the weapon of statistics. Every animal has great volumes of genetic text written in its cells, text most of which, according to the neutralist theory, has nothing to do with fitting it to its peculiar way of life; text that is largely untouched by selection and largely not subject to convergent evolution except as a result of sheer chance. The chance that two large pieces of selectively neutral text could resemble each other by luck can be calculated, and it is very low indeed. Even better, the constant rate of molecular evolution actually lets us date branch points in evolutionary history.

  It is hard to exaggerate the extra power that the new molecular sequence-reading techniques have added to the taxonomist’s armoury. Not all molecular sentences in all animals have yet, of course, been deciphered, but already one can walk into the library and look up the exact word-for-word, letter-for-letter, phraseology of, say, the α-haemoglobin sentences of a dog, a kangaroo, a spiny anteater, a chicken, a viper, a newt, a carp and a human. Not all animals have haemoglobin, but there are other proteins, for instance histones, of which a version exists in every animal and plant, and again many of them can already be looked up in the library. These are not vague measurements of the kind which, like leg length or skull width, might vary with the age and health of the specimen, or even with the eyesight of the measurer. They are precisely worded alternative versions of the same sentence in the same language, which can be placed side by side and compared with each other as minutely and as exactly as a fastidious Greek scholar might compare two parchments of the same Gospel. DNA sequences are the gospel documents of all life, and we have learned to decipher them.

  The basic taxonomists’ assumption is that close cousins will have more similar versions of a particular molecular sentence than more distant cousins. This is called the ‘parsimony principle’. Parsimony is another name for economic meanness. Given a set of animals whose sentences are known, say the eight animals listed in the previous paragraph, our task is to discover which of all the possible tree diagrams linking the eight animals is the most parsimonious. The most parsimonious tree is the tree which is ‘economically meanest’ with its assumptions, in the sense that it assumes the minimum number of word changes in evolution, and the minimum amount of convergence. We are entitled to assume the minimum amount of convergence on grounds of sheer improbability. It is unlikely, especially if much of molecular evolution is neutral, that two unrelated animals would have hit upon exactly the same sequence, word for word, letter for letter.

  There are computational difficulties in trying to look at all possible trees. When there are only three animals to be classified, the number of possible trees is only three: A united with B excluding C; A with C excluding B; and B with C excluding A. You can do the same calculation for larger numbers of animals to be classified, and the number of possible trees rises steeply. When there are only four animals to be considered, the total number of possible trees of cousinship is still manageable at only 15. It doesn’t take the computer long to work out which of the 15 is the most parsimonious. But if there are 20 animals to be considered, I reckon the number of possible trees to be 8,200,794,532,637,891,559,375 (see Figure 9). It has been calculated that the fastest of today’s computers would take 10,000 million years, approximately the age of the universe, to discover the most parsimonious tree for a mere 20 animals. And taxonomists often want to construct trees of more than 20 animals.

  Although molecular taxonomists have been the first to make much of it, this problem of exploding large numbers has actually been lurking all along in non-molecular taxonomy. Nonmolecular taxonomists have simply evaded it by making intuitive guesses. Of all the possible family trees that might be tried, huge numbers of trees can be eliminated immediately — for instance, all those millions of conceivable family trees that place humans closer to earthworms than to chimps. Taxonomists don’t even bother to consider such obviously absurd trees of cousinship, but instead home in on the relatively few trees that do not too drastically violate their preconceptions. This is probably fair, although there is always the danger that the truly most parsimonious tree is one of those that have been thrown out without consideration. Computers, too, can be programmed to take short cuts, and the problem of the exploding large numbers can be mercifully cut down.

  Figure 9

  This family tree is correct. There are 8200794532637891559374 other ways of classifying these 20 organisms, and all of them are wrong.

  Molecular information is so rich that we can do our taxonomy separately, over and over again, for different proteins. We can then use our conclusions, drawn from the study of one molecule, as a check on our conclusions based on the study of another molecule. If we are worried that the story told by one protein molecule is really confounded by convergence, we can immediately check it by looking at another protein molecule. Convergent evolution is really a special kind of coincidence. The thing about coincidences is that, even if they happen once, they are far less likely to happen twice. And even less likely to happen three times. By taking more and more separate protein molecules, we can all but eliminate coincidence.

  For instance, in one study by a group of New Zealand biologists, 11 animals were classified, not once but five times independently, using five different protein molecules. The 11 animals were sheep, rhesus monkey, horse, kangaroo, rat, rabbit, dog, pig, human, cow and chimpanzee. The idea was first to work out a tree of relationships among the 11 animals using one protein. Then see whether you get the same tree of relationships using a different protein. Then do the same for a third, fourth and fifth protein. Theoretically, if evolution were not true for example, it is possible for each of the five proteins to give a completely different tree of ‘relationships’.

  The five protein sequences were all available to be looked up in the library, for all 11 animals. For 11 animals, there are 654,729,075 possible trees of relationships to be considered, and the usual shortcut methods had to be employed. For each of the five protein molecules, the computer printed out the most parsimonious tree of relationship. This gives five independent best guesses as to the true tree of relationships among these 11 animals. The neatest result that we could hope for is that all five estimated trees turn out to be identical. The probability of getting this result by sheer luck is very small indeed: the number has 31 noughts after the decimal point. We should not be surprised if we fail to get agreement quite as perfect as this: a certain amount of convergent evolution and coincidence is only to be expected. But we should be worried if there is not a substantial measure of agreement among the different trees. In fact the five trees turned out to be not quite identical, but they are very similar. All the five molecules agree in placing human, chimp and monkey close to each other, but there are some disagreements over which animal is the next closest to this cluster: haemoglobin B says the dog is; fibrinopeptide B says the rat is; fibrinopeptide A says that a cluster consisting of rat and rabbit is; haemoglobin A says that a cluster consisting of rat, rabbit and dog is.

  We have a definite common ancestor with the dog, and another definite common ancestor with the rat. These two ancestors really existed, at a particular moment in history. One of them has to be more recent than the other, so either haemoglobin B or fibrinopeptide B must be wrong in its estimate of evolutionary relationships. Such minor discrepancies needn’t worry us, as I have said. We expect a certain amount of convergence and coincidence. If we are truly closer to the dog, then this means that we and the rat have converged on one another with respect to our fibrinopeptide B. If we are truly closer to the rat, this means that we and the dog have converged on each other with respect to our haem
oglobin B. We can get an idea of which of these two is the more likely, by looking at yet other molecules. But I shan’t pursue the matter: the point has been made.

  I said that taxonomy was one of the most rancorously ill-tempered of biological fields. Stephen Gould has well characterized it with the phrase ‘names and nastiness’. Taxonomists seem to feel passionately about their schools of thought, in a way that we expect in political science or economics, but not usually in academic science. It is clear that members of a particular school of taxonomy think of themselves as a beleaguered band of brothers, like the early Christians. I first realized this when a taxonomist acquaintance told me, white-faced with dismay, the ‘news’ that So-and-so (the name doesn’t matter) had ‘gone over to the cladists’.

  The following brief account of taxonomic schools of thought will probably annoy some members of those schools, but no more than they habitually infuriate each other so no undue harm will be done. In terms of their fundamental philosophy, taxonomists fall into two main camps. On the one hand there are those that make no bones about the fact that their aim is openly to discover evolutionary relationships. To them (and to me) a good taxonomic tree is a family tree of evolutionary relationships. When you do taxonomy you are using all methods at your disposal to make the best guess you can about the closeness of cousinship of animals to one another. It is hard to find a name for these taxonomists because the obvious name, ‘evolutionary taxonomists’, has been usurped for one particular subschool. They are sometimes called ‘phyleticists’. I have written this chapter, so far, from a phyleticist’s point of view.

  But there are many taxonomists who proceed in a different way, and for quite sensible reasons. Although they are likely to agree that one ultimate aim of doing taxonomy is to make discoveries about evolutionary relationships, they insist on keeping the practice of taxonomy separate from the theory — presumably evolutionary theory — of what has led to the pattern of resemblances. These taxonomists study patterns of resemblances in their own right. They do not prejudge the issue of whether the pattern of resemblances is caused by evolutionary history and whether close resemblance is due to close cousinship. They prefer to construct their taxonomy using the pattern of resemblances alone.

  One advantage of doing this is that, if you have any doubts about the truth of evolution, you can use the pattern of resemblances to test it. If evolution is true, resemblances among animals should follow certain predictable patterns, notably the pattern of hierarchical nesting. If evolution is false, goodness knows what pattern we should expect, but there is no obvious reason to expect a nested hierarchical pattern. If you assume evolution throughout the doing of your taxonomy, this school insists, you can’t then use the results of your taxonomic work to support the truth of evolution: the argument would be circular. This argument would have force if anybody was seriously in doubt about the truth of evolution. Once again, it is hard to find a suitable name for this second school of thought among taxonomists. I shall call them the ‘pure-resemblance measures’.

  The phyleticists, the taxonomists that openly try to discover evolutionary relationships, further split into two schools of thought. These are the cladists, who follow the principles laid down in Willi Hennig’s famous book Phylogenetic Systematics; and the ‘traditional’ evolutionary taxonomists. Cladists are obsessed with branches. For them, the goal of taxonomy is to discover the order in which lineages split from each other in evolutionary time. They don’t care how much, or how little, those lineages have changed since the branch-point. ‘Traditional’ (don’t think of it as a pejorative name) evolutionary taxonomists differ from cladists mainly in that they don’t consider only the branching kind of evolution. They also take account of the total quantity of change that occurs during evolution, not just branching.

  Cladists think in terms of branching trees, right from the outset of their work. They ideally begin by writing down all possible branching trees for the animals they are dealing with (two-way branching trees only, because there are limits to anyone’s patience!). As we saw when discussing molecular taxonomy, this gets difficult if you are trying to classify lots of animals, because the number of possible trees becomes astronomically large. But as we also saw, there are fortunately short cuts and serviceable approximations which mean that this kind of taxonomy can, in practice, be done.

  If, for the sake of argument, we were trying to classify just the three animals squid, herring and human, the only three possible two-way branching trees are the following:

  1. Squid and herring are close to each other, human is the ‘outgroup’.

  2. Human and herring are close to each other, squid is the outgroup.

  3. Squid and human are close to each other, herring is the outgroup.

  Cladists would look at each of the three possible trees in turn, and choose the best tree. How is the best tree recognized? Basically, it is the tree that unites the animals that have the most features in common. We label as the ‘outgroup’ the animal that has fewest features in common with the other two. Of the above list of trees, the second would be preferred, because human and herring share many more features in common with each other than squid and herring do or than squid and human do. Squid is the outgroup because it doesn’t have many features in common with either human or herring.

  Actually, it isn’t quite as simple as just counting features in common, because some kinds of features are deliberately ignored. Cladists want to give special weight to features that are recently evolved. Ancient features that all mammals inherited from the first mammal, for instance, are useless for doing classifications within the mammals. The methods they use for deciding which features are ancient are interesting, but they would take us outside the scope of this book. The main thing to remember at this stage is that, at least in principle, the cladist thinks about all possible bifurcating trees that might unite the set of animals he is dealing with, and tries to choose the one correct tree. And the true cladist makes no bones about the fact that he thinks of the branching trees or ‘cladograms’ as family trees, trees of closeness of evolutionary cousinship.

  If pushed to the extreme, the obsession with branchings alone could give strange results. It is theoretically possible for a species to be identical in every detail to its distant cousins, while being exceedingly different from its closer cousins. For instance, suppose that two very similar fish species, which we can call Jacob and Esau, lived 300 million years ago. Both these species founded dynasties of descendants, which last to the present day. Esau’s descendants stagnated. They went on living in the deep sea but they didn’t evolve. The result is that a modern descendant of Esau is essentially the same as Esau, and is therefore also very like Jacob. Jacob’s descendants evolved and proliferated. They eventually gave rise to all modern mammals. But one lineage of Jacob’s descendants also stagnated in the deep sea, and also leaves modern descendants. These modern descendants are fish that are so similar to Esau’s modern descendants that they are hard to tell apart.

  Now, how shall we classify these animals? The traditional evolutionary taxonomist would recognize the great similarity between the primitive deep-sea descendants of Jacob and Esau, and would classify them together. The strict cladist could not do that. The deep-sea descendants of Jacob, for all that they look just like the deep-sea descendants of Esau, are, nevertheless, closer cousins of mammals. Their common ancestor with mammals lived more recently, even if only slightly more recently, than their common ancestor with Esau’s descendants. Therefore they must be classified together with mammals. This may seem strange, but personally I can treat it with equanimity. It is at least utterly logical and clear. There are indeed virtues in both cladism and traditional evolutionary taxonomy, and I don’t much mind how people classify animals so long as they tell me clearly how they are doing it.

  Turning now to the other major school of thought, the pureresemblance measurers, they again can be split into two sub-schools. Both the sub-schools agree to banish evolution from their day-
to-day thoughts while they do taxonomy. But they disagree over how to proceed in their day-to-day taxonomy. One sub-school among these taxonomists are sometimes called ‘pheneticists’ and sometimes called ‘numerical taxonomists’. I shall call them the ‘average-distance measurers’. The other school of resemblance measurers call themselves ‘transformed cladists’. This is a poor name, since the one thing these people are not is cladists! When Julian Huxley invented the term clade he defined it, clearly and unambiguously, in terms of evolutionary branching and evolutionary ancestry. A clade is the set of all organisms descended from a particular ancestor. Since the main point of ‘transformed cladists’ is to avoid all notions of evolution and of ancestry, they cannot sensibly call themselves cladists. The reason they do so is one of history: they started out as true cladists, and kept some of the methods of cladists while abandoning their fundamental philosophy and rationale. I suppose I have no choice but to call them transformed cladists, although I do so with reluctance.

  The average-distance measurers not only refuse to use evolution in their taxonomy (although they all believe in evolution). They are consistent in that they don’t even assume that the pattern of resemblance will necessarily be a simply branching hierarchy. They try to employ methods that will uncover a hierarchical pattern if one is really there, but not if it isn’t. They try to ask Nature to tell them whether she is really organized hierarchically. This is not an easy task, and it is probably fair to say that methods are not really available for achieving this aim. Nevertheless the aim seem to me to be all of a piece with the laudable one of avoiding preconceptions. Their methods are often rather sophisticated and mathematical, and they are just as suitable for classifying nonliving things, for instance rocks or archaeological relics, as for classifying living organisms.