Climbing Mount Improbable

by Richard Dawkins

  The centipede way of organizing a body is repetitious in a simple way. There is spatial repetition all the way along the train and left-right mirroring within each segment too. But, if we move away from centipedes and their kind, there is a persistent tendency in evolution for segments to become progressively different from each other: not all mutations are simply repeated in every segment. Insects are like centipedes that have lost the legs from all segments except three: segments seven, eight and nine, counting from the front. Spiders have kept legs on four segments. Actually, both spiders and insects have kept more of their primitive limbs’ than this. It is just that they've turned them to other uses, like antennae or jaws. Lobsters and, even more so, crabs, have carried unkaleidoscopic differentiation among segments even further.

  Caterpillars have the usual three ‘proper insect legs’ near the front, but they have also reinvented the leg further back. These reinvented legs are squashier and otherwise rather different from the typical jointed armour legs which sprout from the three thoracic segments. Insects also typically have wings on segments seven and eight. Some {240}

  Figure 7.10 Arthropods are built up from segments repeated, often with variation, from front to back: (from top) mystacocarid crustacean, Demcheilocaris; giant peacock moth caterpillar, Saturnia pyri; dendrobranchiate shrimp, Penaeus; Symphyla (similar to centipedes), Scutigenlla.

  insects have no wings, and their ancestors never had them. Other insects, fleas for instance, and worker ants, have over evolutionary time lost the wings that their ancestors once had. Worker ants have the genetic equipment to grow wings: every worker could have been a queen if she had been reared differently, and queens have wings. Interestingly, a queen usually loses her wings during her own lifetime, sometimes by biting them off herself, when she has completed her mating {241} flight and is ready to settle down underground. Wings get in the way underground, as they do where fleas live, in the thick forest that is their host's fur or feathers.

  Whereas fleas have lost both pairs of wings, flies (there are lots and lots of members of the large fly family, including mosquitoes) have lost one pair of wings and kept the other pair. The second pair of wings survive in greatly reduced form as ‘halteres’, the tiny drumsticks sticking out just behind the working wings (Figure 7.11). You don't need to be an engineer to see that halteres would not work as wings. You need to be quite a good engineer to see what they are actually for. They seem to be tiny stabilizing instruments, doing a similar job for the insect as a gyroscope does for an aeroplane or rocket. The halteres vibrate at the wingbeat frequency. Tiny sensors at the base of the haltere detect turning forces in the three directions known to pilots as pitch, roll and yaw. It is typical of evolution to be opportunistic and make use of what is already there. An engineer designing

  Figure 7.11 All members of the fly family have halteres in place of the second pair of wings. Large flies like these craneflies show them particularly clearly: (left) Tipula maxima; (right) Ctenophora ornata (legs and right wing not shown). {242}

  an aircraft would sit down at a drawing board and design a stabilizing instrument from scratch. Evolution achieves the same result by modifying what is already there, in this case a wing.

  Just having segments evolve to be different from one another is not kaleidoscopic: quite the contrary. But there are other modes of change which we can see as kaleidoscopic, in a more sophisticated sense than we have so far met. Often arthropod bodies have a structure rather like a sentence with rounded off brackets. (If you open brackets in a sentence [inner brackets must be properly nested] the brackets eventually have to be closed again.) The phrase inside the brackets can be expanded or shortened but, no matter how long it is, every ‘(’ must be rounded off with a ‘)’. The same ‘proper nesting’ applies to quotation marks. It also, more interestingly, applies to the embedding of subordinate clauses in sentences. ‘The man who sat on a pin...’ is like an open bracket, demanding a return to the main verb. You can say, ‘The man jumped’, and you can say, ‘The man who sat on a pin jumped’, but you cannot stop at ‘The man who sat on a pin except as the answer to a question or the caption to a picture, in which case the completion of the sentence is implicit. Grammatical sentences demand proper nesting. In a similar way, shrimps, prawns, lobsters and crayfish have six head segments all seized up together at the front end, and at the rear end a special segment called the telson. What happens in between is more variable.

  We have seen one kind of kaleidoscopic mutation, the mirrored mutation reflected about various planes of symmetry. ‘Grammatical’ mutations would be kaleidoscopic in another sense. Once again, the variation that is permitted is constrained, but in this case not by symmetry but by rules such as: ‘No matter how much variation in number of joints you allow in the middle of the leg, the leg must end with a claw.’ Ted Kaehler, of the Apple Computer Company, and I collaborated to write a computer program embodying rules of this kind. It is like the Blind Watchmaker program but the ‘animals’ that it produces are called arthromorphs and their embryology has rules that are missing from biomorph embryology. Computer arthromorphs are a train of segmented bodies like real arthropods. Each segment has a roundish body bit — the exact shape and size is controlled by ‘genes’ like {243} biomorph genes. Each segment may or may not have a jointed leg sticking out each side. That, too, is controlled by genes, and so is the thickness of the leg, the number of joints, the length of each joint and the angle of each joint. There may or may not be a claw at the end of a leg, and that too, together with its shape, is controlled by genes.

  If arthromorphs had the same kind of embryology as biomorphs, there'd be a gene called NSeg which determined the number of segments. NSeg would simply have a value, which could mutate. If NSeg had the value eleven, the animal would have eleven segments. There'd be another gene called NJoint which controlled the number of joints in each limb. No matter how variable they may look — and their variety is my pride and joy — all the biomorphs in the ‘safari park’ of Figure 1.16 have exactly the same number of genes, sixteen. The original biomorphs of The Blind Watchmaker had only nine genes. The colour biomorphs have more genes (thirty-six) and the program had to be completely rewritten to accommodate them. Those are three different programs. Arthromorphs don't work like that. They don't have a fixed repertoire of genes. They have a more flexible genetic system (programming aficionados are the only readers who will wish to know that the genes of an arthromorph are stored as a Linked List with Pointers, while those of a biomorph are stored as a fixed Pascal Record). New genes can spontaneously arise in arthromorph evolution by duplication of old genes. Sometimes genes are duplicated one at a time. Sometimes they are duplicated in hierarchically structured clusters. This means that theoretically a mutant child can have twice as many genes as its parent. When a new gene, or set of genes, appears by duplication, the new genes start out with the same values as the ones from which they were duplicated. Deletion is a possible kind of mutation, as well as duplication, so the number of genes can shrink as well as grow. Duplications and deletions manifest themselves as changes in body form, and are therefore exposed to selection (artificial selection by eye, as for biomorphs). Often a change in the number of genes shows itself as a change in number of segments (Figure 7.12). It can also show itself as a change in the number of joints in a limb. {244}

  Figure 7.12 Arthromorphs differing in numbers of segments. The parent at the top has two mutant offspring.

  In both cases, what emerges is a ‘grammatical’ tendency for middle trucks to drop in and out of the train, while leaving the front and rear trucks intact.

  Duplication or deletion of segments can occur in the middle of an animal, not just at its end. And duplication or deletion of joints can occur in the middle of a limb, not just at its end. This is what gives arthromorph embryology its ‘grammatical’ quality: its ability to delete, or incorporate, the equivalent of a whole relative clause or prepositional clause, in the middle of a larger ‘sentence’. Apa
rt from their property of ‘grammatical’ nesting, arthromorphs have an additional flavour of kaleidoscopic embryology. Each quantitative detail of an arthromorph's body (for example the angle of a given claw, or the trunk width of a given segment) is influenced by three genes which multiply their numerical values together in a way that I'll explain presently. There is a gene specific to the segment concerned, a gene that applies to the whole animal, and a gene that applies to a sub-sequence of segments called a tagma. Tagma (plural tagmata) is a word that comes from real biology. Examples of tagmata in real animals are the thorax and the abdomen of insects.

  For any particular detail, such as claw angle, the three genes that combine to affect it are as follows. First, the gene that is peculiar to the individual segment. This is not kaleidoscopic at all, for when it mutates it affects only the segment in question. Figure 7.13a shows an arthromorph in which every segment has a different value of the {245}

  Figure 7.13 Arthromorphs chosen to illustrate various kinds of genetic effect: (a) arthromorph with different claw-angle gene for every segment; (b) mutation of whole body gene for claw angle; (c) arthromorph with no variation among segments; (d) same as (c) but with a single mutation affecting the whole body gene for claw angle; (e) gradient of segment sizes, not affecting limbs; (f) arthromorph with three tagmata, differing in several features but uniform within each tagma; (g) same as (f) but with a mutation affecting limbs at the level of the (third) tagma; (h) mutation of (f) affecting the limbs in one segment only.

  segment-level gene for claw angle. The result is that each segment has a different claw angle. In all arthromorphs, by the way, there is simple left — right symmetry.

  Moving to the second of the three genes that affects, say, the claw angle, this is the one that influences all the segments of the entire animal. When it mutates, the claws in all the segments simultaneously {246} change, right along the length of the animal. Figure 7.13b shows an arthromorph which is the same as Figure 7.13a except that the claws are slightly pulled in — shortened. The gene affecting claw size at the level of the whole animal has mutated to a smaller value. The result is that the individual claws shrink, while retaining their segment-level peculiarities relative to one another. Mathematically, as I said, this effect is achieved by multiplying the numerical value of each individual segment-level gene for claw angle by the numerical value of the whole body-level gene for claw angle. Claw angle, of course, is only one of many quantitative details being simultaneously determined by similar multiplication sums all along the ‘train. There are whole body genes affecting, say, leg length, and these multiply their values with the segment level genes for leg length. Figure 7.13c and d show arthromorphs which have no variation among segments, but which differ from each other at the level of the whole organism gene for claw angle.

  The third class of gene affects a discrete region of the body, a tagma like the thorax of an insect. Whereas insects have three tagmata, arthromorphs evolve to have any number, and each tagma can have any number of segments: changes in both segment numbers and tagma numbers are themselves subject to mutation in the ‘grammatical’ way we've already discussed. Each tagma has a set of genes that affect the shape of the body and the limbs and claws within that tagma. For example, each tagma has a gene that affects the angle of all the claws within that tagma. Figure 7.13f shows an arthromorph with three tagmata. Most things vary between tagmata more than within any one tagma. The effect is achieved by multiplication of gene values, in the same way as we have already seen for whole body genes.

  To summarize, the final size of each attribute, say claw angle, is arrived at by multiplying the numerical values of three genes: the segment gene for claw angle, the tagma gene for claw angle, and the whole organism gene for claw angle. Since multiplication by zero yields zero, it follows that if, say, the value of a gene for limb size in a given tagma were zero, the segments of that tagma would have no limbs at all, like a wasp's abdominal segments, regardless of the gene {247} values at the other two levels. Figure 7.13g shows a daughter of the arthromorph in Figure 7.13f, which has mutated a gene for limb size at the level of the third tagma. Figure 7.13h is another daughter of the arthromorph of Figure 7.13f, but in this case a limb gene belonging to only one segment has mutated.

  Arthromorphs, then, have a kind of three-tiered kaleidoscopic embryology. They may mutate within one segment, in which case the change is mirrored only once, on the opposite side of the body. They are also kaleidoscopic at the ‘centipede’ or whole-organism level: a mutation at this level is spatially repeated down the segments of the body (and is also left — right mirrored). And they are kaleidoscopic at the intermediate, ‘insect’ or tagma level: a mutation at this level affects all the segments in one local cluster of segments, but not in the rest of the body. I conjecture that, if arthromorphs had to make their living in the real world, their three-tiered kaleidoscopic mutations might have benefits, for the same kind of reasons grounded in evolutionary economics, as we have already discussed in the case of mirrors of symmetry. If, say, the limbs of the middle tagma of the body function as walking legs, while the limbs of the hind tagma of the body function as gills, it makes sense that evolutionary improvements should be repeated serially along the segments of one tagma but not the other: improvements for walking appendages are unlikely to benefit breathing appendages. Hence there may be advantages in possessing a class of mutations which, when they first appear, are already reflected in all the segments of a tagma. On the other hand there may be more particular benefits in making detailed, specialist adjustments to the limbs in particular segments, in which case embryologies may be favoured which have an additional tendency to throw up mutations that are only left — right mirrored. Finally there may sometimes be benefits in mutations simultaneously appearing over all the segments of the body, not completely overriding the existing variation among segments and among tagmata, but weighting them, for example by multiplication.

  As a biologically inspired afterthought, Ted Kaehler and I introduced ‘gradient’ genes into our arthromorph program. A gradient gene sees to it that a particular quality of an arthromorph, such as {248} the claw angle, is not fixed as you move from front to rear of the animal but progressively increases (or decreases). Figure 7.13e shows an arthromorph with no variation among segments apart from a (negative) gradient of segment sizes. The body tapers from front to rear.

  Arthromorphs breed, and evolve by artificial selection, in the same kind of way as biomorphs. A parent arthromorph sits in the centre of the screen, surrounded by its randomly mutated offspring. As in the case of biomorphs, the human selector sees no genes but only their consequences — body shapes — and chooses which will breed (and once again there is no sex). The chosen arthromorph glides to the centre and surrounds itself with a litter of its mutant progeny. As the generations go by, changes in numbers of genes, and changes in values of genes, take place behind the scenes by random mutation. All that the human chooser sees is a gradually evolving sequence of arthromorphs. Just as all computer biomorphs can be said to have been bred from so all arthromorphs can be said to have been bred from . The neat shading of the body segments to make them look solid is a cosmetic touch which does not vary in the existing program, though it could easily be brought under (three-tiered) genetic control in future versions of the program. Comparable to Figure 1.16's biomorph safari park, Figure 7.14 is a zoo of arthromorphs that I have bred from time to time by artificial selection, usually choosing in favour of some sort of biological realism.

  This zoo includes forms that vary at all levels of the kaleidoscopic embryology. You can recognize creatures with tapering bodies as having at least one gradient gene. You can recognize clear-cut division into tagmata: groups of neighbouring segments resemble each other more than they resemble other segments. But you can still recognize some variation in form even among the segments within a tagma. Real insects, crustaceans and spiders vary in similarly tiered kaleidoscopic ways. Especially revealing are the so-called hom
eotic mutations of real arthropods, mutations that cause one segment to change so that it follows the pattern of development normal for a different segment.

  Figure 7.15 shows examples of so-called homeotic mutations in the fruitfly Drosophila and in the silkworm caterpillar. The normal Drosophila, like all flies, has only a single pair of wings. The second {249}

  Figure 7.14 Arthromorph zoo. A collection of arthromorphs bred by artificial selection with an eye to their resemblance, however vague, to real arthropods. {250}

  Figure 7.15 Homeotic mutations: (a) four-winged Drosophila. In normal Drosophila the second pair of wings is replaced by halteres, as in Figure 7.11; (b) normal (upper) and mutant (lower) silkworm caterpillars. Normally there are proper insect legs only on the three thoracic segments. The mutant has nine ‘thoracic’ segments.

  pair of wings is replaced by halteres as explained above. The picture shows a mutant Drosophila in which not only is there a second pair of wings instead of halteres, the entire second thoracic segment is reduplicated in substitution for the third thoracic segment. In arthromorphs this effect would be achieved by a ‘grammatical’ duplication followed by deletion. Figure 7.15b shows a mutant silkworm caterpillar. Normal caterpillars have three ‘proper’ jointed legs like any other insect, although, as I said, their rear segments have squashier, ‘reinvented’ legs. But the mutant caterpillar at the bottom of Figure 7.15 has nine pairs of ‘proper’ jointed legs. What has happened is that segments of the thoracic tagma have been duplicated, {251} just as in the right-hand arthromorph of Figure 7.12. The most famous homeotic mutation is ‘antennapedia’ in Drosophila fruitflies. Flies with this mutation have a normal-looking leg poking out of the socket where an antenna ought to be. The legproducing machinery has been switched on in the wrong segment.