Nursery rhymes can be unspeakably violent. Recall, for example, the three unfortunate mice who, as well as being blind, had their tails docked by a sadistic, knife-wielding peasant. The real-life, adults-only version of this tale features a newborn mouse, seemingly perfect in every way except that the front end of its body is oddly truncated, ending in a small mound crowned by two tiny ears. This mouse is certainly blind, since not only does it have no eyes, it also has no head. It looks as cleanly decapitated as if by the farmer’s wife of folklore. The mouse was one of four born without a head, as a result of a mutation in a regulatory gene called Lim 1. With no mouth or nose to breathe through, the mouth died very soon after its birth. More than a hundred other headless mice foetuses did not get as far as being born.
The experiments in which these mice were created1 were part of an effort to understand the activities of Lim 1, one of an increasingly well-documented cadre of regulatory genes whose role it is to ensure that every part of the body develops in the place it should. When such genes are mutated, the result can be as monstrous as anything from nursery folklore. The work on the headless mice is just one of many examples of studies of genes whose action goes well beyond the maintenance of what we would consider the normal range of variation. The genetic variation that dictates whether your eyes are brown or blue is superficial – but there is a more fundamental kind of variation, in genes that determine whether you have a head at all for eyes to stare from.
Lim 1 belongs to a group of regulatory genes called Hox genes, and their story really begins with Bateson’s descriptions, in Materials for the Study of Variation, of insects with the antennapedia mutation in which tiny simulacra of legs grew out of the head in place of antennae; of insects with extra wings; and of people in which vertebrae in one part of the spinal column had been transformed into vertebrae characteristic of an adjacent region of the spine. Bateson realized that all these syndromes were characterized by a kind of abrupt discontinuity in which body parts developed in places that would normally be occupied by others, but with the rest of the body eerily undisturbed. He called this phenomenon homeosis, and the genes responsible for such mutations became known as homeotic genes.
Decades later, a family of homeotic genes were each found to have a very characteristic region of their DNA, 180 bases long, which, when translated into the amino-acid sequence of a protein, made a kind of prong which fitted neatly onto the operator regions of other genes. When geneticists wish to pinpoint specific parts of a DNA sequence on a computer printout, they often draw a rectangle or ‘box’ around it. The DNA-binding region of homeotic genes therefore became known as the ‘homeo-box’,2 later contracted to Hox. Other families of genes have since been discovered, each with their own distinctive DNA-binding sequences. One such gene was found to be responsible for a mutation in Drosophila known as paired, in which the segments of the embryo failed to develop properly. This gene contains a characteristic sequence known as the paired box, and a family of related genes has since been found in many other animals.3 This family has acquired the name of Pax genes, and its most famous member – Pax 6 – is involved in the development of eyes throughout the animal kingdom. Mutations in Pax 6 turn out to be implicated in congenital eye defects in humans as well as in mice and flies, and the artificial expression of Pax 6 in flies outside the head has created bizarre mutant flies in which eye tissue appears all over the body. Pax6 demonstrates a startling coherence of plan which unites otherwise diverse animals. (One cannot help feeling that Goethe and Geoffroy would have approved.) But whether they are called Hox or Pax, such families of genes clearly encode proteins similar in concept to the lac or lambda repressors – that is, they are regulators.
Bateson advanced homeosis as just that kind of discontinuous variation that might stand in opposition to the ‘insensible gradation’ of Darwinian change. He thought that such clear-cut discontinuities were a feature of all genetic variation. In that, he was mistaken. However, what he had discovered was a stratum of genetic variation beneath the kind of variation studied by the biometricians – a genetic substructure over which the everyday kind of normal variation is draped, and on which it depends for its expression.
Such variation is, almost by definition, hidden, and very hard to reach in conventional breeding experiments. The shock value of monsters is a function of their rarity: because monstrosities involve major disruption to body form, it is rare that they live long enough to be born, let alone be capable of reaching adulthood and taking part in the kinds of breeding experiment directed by Mendel or Morgan. Headless mice can be born only because of careful breeding of a stock of mice, each of which carries one recessive, mutant allele of the Lim 1 gene. Such a mouse will appear normal because the decapitating effects of the recessive, mutant allele will be masked by the dominant, normal allele on the accompanying chromosome – one normal allele out of the pair is all that is required to get a head. Only when these apparently normal mice are interbred is there any chance of producing headless offspring. Because the headless form of the mouse depends on the conjunction of two recessive Lim 1 alleles, potentially headless embryos will be produced in the classic ratio of three normal offspring for every one that is headless, just like Mendel’s dwarf pea plants. In the case of the headless mice, however – unlike the dwarf peas – very few survive long enough to be born and have their condition reported.
The rarity of recessive monsters was just one of the problems that confronted Edward B. Lewis, a pioneer of research into what became known as the Hox family of genes, who started in Morgan’s footsteps by breeding flies – but with the ambitious aim of mapping the genes that, when mutated, created the whole homeotic freak show. Lewis joined Caltech’s successor to Columbia’s Fly Room, obtaining his doctorate there in 1942, and was especially interested in an extravagant homeotic mutation called bithorax. The middle section of the fly, behind the head and in front of the characteristically limbless abdomen, is called the thorax. This is the fly’s engine room, carrying everything it needs to move from place to place. Each of the three thoracic segments bears a pair of legs, and the second segment a single pair of wings. In insect terms, flies are highly evolved – more primitive insects, such as dragonflies, have two pairs of wings. In flies, the second pair of wings – sprouting from the third thoracic segment – has evolved into a pair of drumstick-shaped structures called halteres. These are organs of balance, which whirl around like gyroscopes and help make flies such expert aeronauts. Flies with the bithorax mutation are hard to miss because they have not one but two pairs of wings: one pair growing from the second segment, as normal, and another emerging from the third. A closer look reveals that this mutation is more than just the addition of wings. The entire third thoracic segment, rather than developing as normal, has developed as a duplicate of the second segment.
With immense care and decades of patience borne out of thousands of breeding experiments, Lewis found that the gene responsible (named bithorax, after the mutation resulting from defects in the gene) was part of a closely linked cluster of homeotic genes involved in specifying the structure of the thorax and abdomen of the fly. This cluster was similar to an operon, except that all the genes involved were regulators. Indeed, some of them appeared to regulate themselves as well as one another, offering the possibility of great control over detailed structure. Another cluster of homeotic genes was found to be involved in the structure of the head end of the fly. This cluster was the ‘ANT-P’ or antennapedia complex, containing the antennapedia gene responsible for the monsters that Bateson had noticed and catalogued, those insects with legs growing from their head in place of antennae.
It turned out that the fly is unusual in that what was once a single cluster of genes has been split into two: the antennapedia and bithorax complexes. This may be a peculiarity of flies and their close relatives. In most animals more complex than jellyfish, there is single cluster – the Hox cluster – which is equivalent to the two clusters in flies. This cluster can be thought of as a single operational unit whose members are responsible for ensuring that each segment of the animal develops the appropriate structures. Failure to do this results in homeosis, in which structures are matched with the wrong segments – legs on the head, for example.
Lewis discovered a particularly striking feature of the Hox cluster: the genes are arranged within the cluster in the same linear order as the segments in the animal whose identity they govern. For example, the Hox genes responsible for the head end sit next to the genes responsible for the thorax, and further along the cluster are genes responsible for setting the identity of progressively posterior parts of the abdomen.4 This remarkable feature, known as collinearity, can hardly be a coincidence. The Hox genes are of such importance in creating the shape of an animal that close and coordinated regulation is essential. As many Hox genes regulate one another, it makes intuitive sense for a gene involved in specifying a particular part of the body to be more concerned with genes involved with the formation of adjacent regions, rather than with ones farther away.
This close regulation presumably explains why Hox clusters have been found in all animals with a distinct front and rear end. Animals as diverse as mice, humans, flies and worms have at least one cluster of Hox genes in which the genes are arranged in an order corresponding to the position of the structures they specify. Simple animals such as jellyfishes and Hydra (of regenerative notoriety) also appear to have Hox genes arranged in a linear order. Sponges, the simplest multicellular animals, appear to have isolated Hox genes, but not, it seems, a linear cluster of them – and this may be connected with their lack of any regular body patterning above the level of simple tissues. Single-celled creatures such as yeasts also have Hox genes, although, as in sponges, they are not in clusters. Plants do not appear to have Hox genes closely related to those found in animals, probably because plants do not organize their bodies in the same stereotyped and tightly organized way that is characteristic of animals. Plants do, however, have other homeotic genes, mutations of which lead to the kinds of monstrosity – petals developing as leaves, and so on – that led Goethe to suggest that plant structures have a unity of plan in which all organs can be seen as elaborations on an archetype of the leaf.
The closeness with which Hox genes regulate one another is illustrated – startlingly – by the fact that Hox genes from one species can regulate corresponding Hox genes in another, totally different species. A Hox gene called deformed, for example, is involved in shaping the back of the head in flies. In addition to the DNA sequence that encodes a regulatory protein, the deformed gene contains a regulatory element akin to the operator sequence that Jacob and Monod found in the lac operon. This element is essential for the deformed gene to do its job properly. Mammals, such as mice and humans, also contain a gene very like deformed which is involved in the development of the back of the head, specifically part of the hindbrain. It is reasonable to suppose that the mammalian and fly versions of deformed are homologues – that they share a common evolutionary descent from a gene which shaped the backs of the heads of the long-extinct common ancestor of flies and mammals some 600 million years ago.
Given that all other traces of this common ancestor vanished more than half a billion years ago, it is remarkable that genes from flies and mammals can stand in for each other. In 1992, Alexander Awgulewitsch and Donna Jacobs of the Medical University of South Carolina took the deformed regulatory element from flies and showed that it regulated the activity of the mouse version of deformed, in the mouse hindbrain. To show that this was no fluke, William McGinnis of Yale University and colleagues did a converse experiment, showing how regulatory elements associated with the version of deformed found in humans could substitute for the deformed regulatory element in embryonic flies.5 These amazing results show that Hox genes have been as vital to the evolution of animal form as they are to the creation of every new animal embryo – whether fly maggot or human baby.
Had a messenger arrived at the Académie des Sciences in the spring of 1830, bearing a message from the future pointing out these remarkable similarities in the genetic substructures of animals, Geoffroy would have cheered, Cuvier would have gnashed his teeth in impotent rage, and Goethe would have smiled in quiet satisfaction. It does beg the question of why, given these similarities, animals come to look as different as they do. Hox genes or not, maggots remain maggots, babies are still babies, and Jeff Goldblum will never really make that Kafkaesque metamorphosis into a giant insect.
Some of the differences are related to the numbers of Hox clusters in a given genome. Although most animals have just one Hox cluster, the genome of vertebrates has been duplicated at least once, leading to the wholesale duplication of Hox genes. So whereas flies have a single Hox cluster, mammals such as mice and humans have four. Although a Hox gene in one cluster may correspond to genes in the other three clusters, the duplication has created the possibility that each new gene can evolve in its own way, giving immense scope for greater complexity and refinement in gene regulation. It is possible that the great structural complexity of vertebrates compared with other animals results, at least in part, from this multiplicity of Hox genes.
In 1994 this possibility was thrown into sharp relief by the discovery, by the British zoologist Peter Holland and the Spanish researcher Jordi Garcia-Fernandez, that lancelets have a single Hox cluster.6 Given that these vaguely fish-like animals are the closest relatives of the vertebrates, but lack quintessentially vertebrate attributes such as the neural crest, it is possible that we owe our brains, skulls, teeth, sense organs, limbs – indeed, everything in which neural crest tissue plays a part – to a duplication of a primordial Hox cluster in our fish-like ancestors more than 550 million years ago.
But lest we are tempted to use Hox cluster duplication as a way to justify our self-appointed station at the pinnacle of creation, we should note that some vertebrates have gone even further. The most diverse and successful group of vertebrates by far is the teleost fishes, encompassing virtually all familiar species of bony fishes (and many unfamiliar ones). There are more species of teleost than of all other species of vertebrate – other fishes, birds, amphibians and mammals – put together, and they are found in a great diversity of form, ranging from tuna to seahorses. The reason for this great diversity and pronounced success may lie in Hox cluster duplication: some teleosts appear to have no fewer than eight Hox clusters. Even the humble zebra fish (Danio rerio), stalwart of aquariums in the waiting rooms of a million dentists, is known to have seven Hox clusters.7
Other large-scale differences in animal form are undoubtedly accounted for by other differences in gene structure, less obviously dramatic than bulk duplication. Although Hox genes in a fly are homologous with those in a human, they differ in a host of details as regards sequences and structures. For example, the Drosophila cluster has relatively small numbers of rather extensive genes, whereas each human Hox cluster contains many more genes, but physically smaller – an entire human Hox cluster could fit into the space occupied (in terms of bases of DNA) by antennapedia, just one of the genes in the Hox cluster in fruit flies. Such differences, accumulated over history, might ultimately contribute to differences in the physical appearances of animals.
In any case, humans and flies are so divergent, in terms of their evolutionary histories, that it is hard to be certain about how they got that way, relative to each other. Comparisons between more closely related creatures reveals a clearer answer – that animals differ in appearance not so much because of the accumulation of mutations in structural genes (although that is surely important), but because of evolutionary changes in the way that genes are regulated. So, while flies are very different from humans, they are much less different from other jointed-legged animals, such as shrimp. Work on Hox genes in shrimp and other crustaceans has begun to reveal how changes in regulation have had an impact on evolution. To that extent, Cuvier was right to confine comparison within strictly defined embranchements.
Flies and shrimp are both arthropods, animals whose armoured bodies are arranged in a clearly defined series of segments. Arthropods tend to be classified according to how their segments are disposed. Insects, for example, are very neat – the head region contains a number of mouth parts, and the three segments of the thorax each contain a pair of legs. The second – and sometimes the third – thoracic segments bear a pair of wings. None of the several segments in the abdomen bear limbs or any other kind of appendage. Insect limbs are never branched, and tend to be very specialized. An appendage used for feeding will not be used as a walking leg, or vice versa. Crustaceans, though, have many more, and more diverse, appendages than do insects, and their appendages can be found all along the body, from head to tail. They are often branched and multifunctional, with different parts of the same appendages used for walking, swimming, eating or a combination of all three. In addition, some appendages carry gills for breathing.
Despite these obvious differences, insects and crustaceans are closely related. Some scientists would go further and say that insects evolved from a particular crustacean stock. In other words, insects are a group of crustaceans that specialized in life on land, in the same way that land vertebrates (amphibians, reptiles, mammals and birds) are land-living fishes. This implies that the concise body plan of insects evolved by a process of simplification of the more extravagant crustacean form. Crustacean gills would be useless for breathing on land, and insects have evolved a new system of air-tubes, unconnected with their limbs. Some scientists have speculated that insects transformed the gills of their crustacean ancestor into wings – and that the first insects to fly had many more than one or two pairs of wings. Such engaging ideas aside, it is abundantly clear that crustaceans have many legs on the abdomen – whereas insects have none in this region.
The reason for this difference lies entirely with the evolution of gene regulation. The presence of appendages on the abdomen is determined by a regulatory mechanism involving a Hox gene called Ultrabithorax. In crustaceans, this mechanism ensures that appendages appear on the abdomen in their appropriate positions. In insects, however, the same mechanism is tuned such that the formation of appendages is suppressed. Insects and crustaceans are sufficiently similar for us to be sure that they both have Ultrabithorax genes, and that these genes will be regulated in networks which bear far closer comparison than would be the case with, say, the deformed gene in flies and mice.8 It is thus possible to see how the different forms of flies and shrimp have been shaped by evolutionary changes in gene regulation.
For all these insights – the connection between homeosis and Hox genes, the organization of Hox genes into clusters, and the implications of this clustering for our understanding of the origin of form – we owe an enormous debt to the dogged persistence of Edward Lewis, finally rewarded in 1995 with a Nobel prize. He shared the award with two other scientists who delved even further into the hidden realm of genetic regulation – deeper even than Hox genes.
Hox genes are remarkable morphological generators, but they are not the only game in town: they comprise just one relatively superficial level in a multi-tier system of genetic regulation. These further, occult substrata of regulation were explored in a series of elegant experiments on Drosophila embryology, published in 1980 by the German scientist Christiane Nüsslein-Volhard and the American-born Eric Wieschaus, then both at the European Molecular Biology Laboratory in Heidelberg, Germany. Their mission was clear: to make a complete catalogue of monsters which they would use to systematize our understanding of how a fly develops from an egg.9 As such, their work would fulfil the programme set out by Bateson in Materials almost a century earlier.
In a fruit-fly breeding programme of epic scale, Nüsslein-Volhard and Wieschaus set out to discover which of the approximately 20,000 genes in the Drosophila genome were involved in development.10 They discovered a select group of mutations which affected fruit-fly morphology – or rather, the form of fruit-fly maggots, given that most of the mutations were so severe that development beyond a very early stage would have been impossible.
It was evident that Hox genes acted in embryonic development at a relatively late stage. Before you can create a homeotic mutant in which legs are traded for antennae, it is first necessary to be able to identify legs and antennae as separate and distinct structures, within the context of a body plan which already exists. Hox genes are important, but in the great scheme of things they are the interior decorators that move in to paint an otherwise finished house.
Some mutations have far greater effects, but they are not readily apparent in breeding experiments: like the headless mice, the afflicted embryos are so monstrous that they do not develop very far before dying. For example, there is a mutation in flies called bicaudal that creates an embryo with two rear ends stuck together, with no head or thorax. This kind of mutation, resulting in global disruption rather than tweaks in this segment or that, represents a seismic disturbance in the deepest layer of regulation. Nüsslein-Volhard and Wieschaus studied a range of such mutations – grubs with whole swathes of adjacent segments missing; or with only odd-numbered or even-numbered segments missing; or with all the segments present, but the structures within them arranged back to front. By examining and classifying these monsters in a manner befitting Bacon or Pare, the scientists could, at last, piece together the story of how a fly develops from egg to embryo, in terms of genetic regulation.
The most profound level of genetic regulation – and the one that takes place at the earliest stages – relies not on the embryo’s own genes but on those of its mother. Proteins produced by the so-called maternal-effect genes map out a global orientation for the fertilized egg, laying down which end will be the head end, and which the rear. It is at this level that mutations such as bicaudal act. Once the overall plan is sketched out, a set of the embryo’s own genes start dividing the embryo into large but well-defined regions. Mutations in these gap genes result in the absence of several adjacent segments – the entire thorax and most of the abdomen, say, for mutations in the gene Krüppel. Only when the body has been divided into these regions can the individual segments be created. The regulation of this segmentation process depends strongly on the activities of the pair-rule genes that determine alternating segments (paired is one of the genes in this class). Mutations in a pair-rule gene called even-skipped result in an embryo that lacks even-numbered segments. For odd-skipped, the converse is true. The fourth layer of regulation, above maternal-effect, gap and pair-rule genes – is ruled by the segment-polarity genes, which determine the positions of structures in individual segments. Mutations in these genes are more subtle than the others, and can usually be discovered only by careful study of superficial anatomical features.
Only then, once the body is divided into a back and a front, into large regions, and into segments which are in the correct orientation, can Hox genes have a stage on which to act. Because they play their parts when much of the important action has already taken place, animals with mutations in Hox genes often have a chance of surviving until adulthood – allowing them, in some cases, to survive in the wild so that they can come to the attention of naturalists such as Bateson.
In their revelation of a previously hidden wealth of developmental mutations, Nüsslein-Volhard and Wieschaus presented a show of grotesques that no medieval bestiary could possibly have matched. More seriously, they completed the systematic investigation of the causes of monstrosity started so long before by Paré and continued by Bateson. At a more human level, their results have dramatic implications for human reproduction, and set one aspect of human misery in context. It is thought that a large percentage of early miscarriages – around 40 per cent – may be caused by mutations that affect the fundamental organization of the body. Indeed, mutations that produce gross abnormalities of form will cause spontaneous abortion very early on in pregnancy, often before a woman realizes that she might be pregnant. The mutations studied by Nüsslein-Volhard and Wieschaus were very much of this kind: mutations that killed a fly long before it grew out of the grub stage.
It would be wrong to put too great an emphasis on a picture of genetic regulation as a series of layers, each self-sufficient and discrete but for commands fed to a layer above, with the uppermost layer – the one that is easily visible, and what we would regard as the source of normal variation – constituting the most trivial, at least in respect of the origin of form. Some apparently superficial variation provides a window on to deeper processes. One case involves a gene called bric-a-brac that specifies how the pigmentation of the fruit-fly abdomen differs between male and female flies – as superficial a function as might be imagined. But bric-a-brac is part of a regulatory network which includes Hox genes, and may even have evolved from a homeotic gene itself. On the other hand, the activities of bric-a-brac today regulate how flies choose their mates, thus making a connection between the present-day courtship behaviour of flies and genes that control the origin of fruit-fly form.11
In another case, the protein products of genes called engrailed and wingless play an important part in ensuring that the segments of the developing grub are properly oriented. But each of these genes plays other roles at different times during the developmental process. As its name suggests, wingless has to do with the specification of wings; mild mutations in the gene result in wingless flies. The name engrailed is more obscure. The term comes from heraldry and first appears in this context in the Morte D ‘Arthur, the canonical retelling of the King Arthur legends, in the early fifteenth century. It refers to the condition of edges or borders ornamented with semicircular indentations – a reference to the appearance of segmental boundaries in mutant flies. But in addition to its role in segmentation, engrailed plays a part in the development of nerve cells in most, if not all, animals at various stages of their development. Indeed, many scientists believe that the role of engrailed in the formation of nervous systems is its primary, even primordial function, and only later was it co-opted into the network of genes specifying segmentation. In this way, genes involved in one regulatory layer may become involved in others, at a different time in individual development, creating a rich network of connections which blurs the edges between individual layers in the hierarchy.
The efforts of Lewis, Nüsslein-Volhard, Wieschaus, and many others have given us a rough understanding of the layout of genes in a network, creating sketch-maps of the origin of form. The next step will be to make those maps come alive: to investigate how such networks behave as discrete entities in their own right rather than as sums of their genetic components. This is important because networks may have properties all their own which could never be guessed by looking at lists of genes – properties which might shed light on many things that have remained obscure. They may help us to see why it is that monstrosities can distort some parts of the body beyond recognition without affecting other parts; whether the differences between everyday variation and monstrosity are just differences in degree, or reflect a qualitative shift; why variation occurs at all; and how genetic networks can create babies with such startling fidelity and reliability – questions which have perplexed thinkers for millennia.
Here the story of the genome turns from history and becomes reportage. The application of information theory to genetics is a relatively new but rapidly expanding discipline. In the past few years, terms such as ‘bioinformatics’, ‘systems biology’ and ‘computational biology’ have entered the scientific lexicon.
One team of computational biologists, Garrett M. Odell and his colleagues at the University of Washington in Seattle, has been using the network approach in an investigation of engrailed, wingless and other segment-polarity genes. Taken together, these genes form a small network that ensures that structures in the developing segments of fruit-fly grubs appear in the correct front-to-back order. This ‘segment-polarity module’ is a semi-autonomous subroutine of the greater network that comprises the entire genome. By studying the module as an integrated entity, rather than focusing on any particular gene or interaction, the researchers hope to learn how genetic networks (and, by extension, whole genomes) behave in general. It is surely too early to put such cutting-edge research in its proper historical context. But what makes the Seattle group’s otherwise terse and technical report so engaging is that they clearly knew they were on the verge of important discoveries – knowledge that fuelled a determination to continue despite formidable obstacles along the way.12
The scientists’ aim was to create a mathematical model of the segment-polarity module, run the program, and see if the computer equivalent of neat, stable segmentation would emerge, in imitation of what actually happens when genes such as engrailed and wingless interact to order the segments of a fly. That way, they could test the reaction of the network to various disturbances and gain an insight into how networks behave without having to embark on tens of thousands of breeding experiments – potentially collapsing decades of work into a matter of weeks or months. This may sound simple, but the quest to make a functioning virtual segment-polarity network soon ran into serious difficulties.
The first task was to collect everything that had ever been discovered about the interaction between the genes in the segment-polarity module, and assemble these facts into a kind of genetic circuit diagram. In the fly, genes in the segmentation module have rather quaint names. Apart from engrailed and wingless, there is hedgehog, patched and cubitus interruptus – names derived from the appearances of flies with mild mutations in the genes concerned. But making a network of genetic interactions requires more than drawing arrows between words such as hedgehog and patched. More information is needed, specifically about the strengths of various interactions, such as how tightly a regulatory protein binds to an operator’s DNA, under what conditions and for how long. But our knowledge of such things is very patchy indeed: because regulatory proteins are found in much smaller quantities than are the structural proteins whose abundance they regulate, it is extremely difficult to measure their properties accurately. In any case, most biologists are more interested in the mere fact of the interaction between gene and regulator than in the minutiae of how such interactions are governed at a biochemical level. In the absence of real, known values for these necessary quantities, the researchers labelled each value as an unknown ‘free parameter’ – an x in the equation. The first essay at a network had no fewer than fifty such unknowns, reflecting our ignorance of the fine details of a system even as well known as the fly segment-polarity network, and making the creation of even a toy version of any genetic network a remote prospect indeed.
At this point, Odell’s team hit on a way to answer the question, but from a different angle. Given that we do not know the actual values of the unknowns, they asked, might it nevertheless be possible to devise some plausible set of values for these unknowns that would make the virtual segment-polarity model work? The answer was a decisive ‘no’. The researchers generated thousands of sets of values for the unknowns, fed them into their model and ran the program. The results were not encouraging. In most of the runs, the virtual network never settled down to a state that was stable for long enough to allow segments to form. For those few sets of parameters that produced a stable system, some of the virtual genes were switched on all the time, while others were continually repressed – reminiscent of the mutations in the lac repressor or operator of E. coli that led to the continual synthesis of beta-galactosidase. In only one run in 3,000 did anything like stable segmentation appear, and only if parameters were tuned very precisely. If just one of the parameters varied by a small amount from its optimum value, the whole system would collapse.
Struck down not once but twice, the scientists now had what proved to be a remarkable insight. They knew that in real life a segment-polarity module must exist, and that it is stable and robust enough to work beautifully in an enormous variety of otherwise quite different creatures. In the language of computer science, a segment-polarity model exists that produces stable solutions when offered an extremely wide range of parameter values. If this were not so, we should not see the results of its work all around us. As Goethe, Geoffroy and Bateson had realized, bodies are often made of repeated segments, whether manifested in the segments of insects or worms, the vertebrae of backbones or even the arrangement of leaves on stems – all suggestive of an underlying natural architectural preference for modular construction. Scientists later confirmed and emphasized this universality, by showing that the same segment-polarity genes that orient the segments in insects organize the development, in discrete compartments, of the brain of human embryos. Segmentation was clearly out there in nature, a production of a network of interacting genes, and was therefore something that could in principle be modelled on a computer. The problem lay not with the approach as a whole, but in the particular model the scientists were using, which was not capable of producing the stable, robust solutions seen in nature. Something was wrong with the model.
It turned out that just two connections had to be added to the circuit diagram of the network to make it function properly. In one, cubitus interruptus had to repress engrailed. In the other, wingless had to activate itself. Crucially, the researchers knew of no biochemical evidence to substantiate these links. All they knew was that these links had to be present for the model to work as a whole. At this point, the researchers crossed an important philosophical divide: from biology as a description of nature to biology as a body of predictive theory. Rather than simply collecting data on how molecules interact – which is really a kind of natural history, albeit on a molecular scale – they were able to use their model as a way of predicting discovery, even directing it. This way of doing science transcends the particulars of biology to become a form of physics, like the particle physics of the mid-twentieth century in which theorists predicted the existence of hosts of the exotic particles they needed to make their picture of the Universe hold together – and experimenters went on to find that the theorists had been right. It is its predictive quality that makes the Seattle group’s work a milestone. Happily, the researchers found – after the event – reports describing evidence for these two missing connections.
Once Odell’s team added the two links to their model, success was instant and complete. Stable patterns of segmentation seemed to burst from the computer, almost irrespective of the values chosen for the unknowns (reduced from fifty to forty-eight in the revised model). From 240,000 trials, 1,192 – around one in 200 – produced a stable, realistic segmentation pattern, even though each parameter could take any of a huge range of values. At first sight, a success rate of one in 200 seems rather poor, but this should be seen in the context of the huge variability of the parameter values. The researchers calculated that if the model allowed the values of each of the unknowns to vary over just 10 per cent of a hundred- or thousand-fold range, a random search would come up with not one solution in 200 trials, but one in 1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000. The implication is that parameters in the network can vary by very much more than 10 per cent in value, and the network will still produce stable segmentation patterns.
In engineering terms, even a 10 per cent variation represents unbelievable sloppiness. Imagine a factory turning out toasters, each one made from fifty parts, and that the quality control department allows for a 10 per cent error on every part. If such a production line operated for a time equal to the age of the Universe, no functional toaster would ever roll off the end. But put the segment-polarity network in charge, and a perfectly good toaster would made in every few hundred attempts, given the same or even a greater error rate. Clearly, in the real life of genes and proteins, there is no such thing as a ‘best’ set of parameters: the segment-polarity network is constructed in such a way that an enormous range of values in the unknowns will produce a stable pattern of segmentation.
Odell’s group looked at one of the unknowns, the rate at which the protein product of the wingless gene diffuses through a cell. Nobody knows the real value of this rate, but that turned out not to matter very much: the model yielded stable solutions when the protein diffused rapidly across cells, and stable solutions when there was hardly any diffusion at all. The model as a whole proved itself to be thoroughly robust, provided the network was wired up correctly. This aspect is the key. The precise base sequence of genes is of little consequence as long as the gene works after some fashion, because the network tolerates huge variations in the efficiency of individual components.
The robustness that Odell and colleagues found in their model network reflects a reality of development that has hitherto remained unexplained in anything more than vague terms. A robust network is defined as one that produces workable results given a wide range of input values.It can tolerate sudden change, and can absorb great damage, yet still produce a viable organism. A good example of a robust network is the Internet, originally designed by the military as a distributed system of computers that would be resistant to large-scale attack. Damage to any one of the Internet’s nodes is quickly countered by routing information through other nodes. The system as a whole is therefore able to sustain surprisingly large insults without catastrophic loss of function. Crucially, the function of the Internet is a matter of connectedness, and is not limited by the specifications of any component part – all you need is a modem, and you can connect your basic home computer to a network shared by the most powerful machines ever invented.13
Genomic networks, like the Internet, are robust in that they can tolerate enormous flexibility in the efficiency of their various components. They also resemble the Internet in that they are distributed – there is no single, vulnerable hub through which all information must pass. As we have seen, there are subroutines, or modules, in which a set of genes is clearly more connected to others in the same subset than with genes more generally. This makes networks tractable for study, because it is possible to examine modules more or less in isolation – the Seattle group’s work on the segment-polarity network is a case in point. But modularity is a desirable property of robust networks, because damage to any one part of the network can be contained relatively easily, without compromising the rest of the network.
Modularity explains why it is possible to produce dramatic monstrosities in which one part of an animal is distorted, duplicated, substituted for another part – or even completely absent – while other parts develop quite normally. Lim 1-mutant mice have no heads, but their limbs, tail and bodies are normal, at least from a superficial inspection.14 In the classic antennapedia mutation as studied by Bateson in insects caught in the wild, the monstrosity does not spread beyond the leg-like antennae, and the affected insect is otherwise quite normal. This modular damage-limitation strategy is evident even with the more drastic mutations described by Nüsslein-Volhard and Wieschaus. A mutation in a pair-rule gene such as even-skipped may cause the deletion of even-numbered segments, but the odd-numbered segments form regardless. Likewise, a mutation in a gap gene such as Krüppel may result in the deletion of large regions of the body, but segments still form in the parts that remain. That is, mutations in such genes do not result in the total collapse of the network, creating embryos which are, invariably, formless blobs of protoplasm. The modular construction of networks ensures that the unaffected parts of the body develop normally, as if in complete isolation from the total devastation just a few cells’ breadth away.
The toleration of networks for enormous ranges in the efficiency of their individual components could also help explain the nature and existence of variation. Darwin was the first to understand the crucial importance of variation to evolutionary change. As such, he was keenly aware that his theory of natural selection was utterly dependent on the existence of variation, and that natural selection could not function without some mechanism to replace the variation culled in each generation by selection’s scythe. Darwin’s failure to advance such a mechanism contributed to the eclipse of natural selection at the end of the nineteenth century, creating a void that could be filled only with the kind of vacuous evolutionary storytelling that drove Bateson and Morgan towards a more experimental approach to the problem. Nobody can doubt the importance of what happened next: the subsequent advances of genetics, the chromosome theory, the rehabilitation of Darwinism in the ‘modern synthesis’, the discovery of the structure of DNA, and so on were a direct result of the dissatisfaction so eloquently expressed by Bateson in Materials for the Study of Variation. As a result, we are now able to study variation in all its forms, from individuals in populations down to the interactions of genes and proteins.
Given the (rightful) domination of evolution by natural selection as the pre-eminent theory of why the natural world is the way it is, it is all too easy to explain the existence of variation only in terms of its contribution to evolution. Mutations, whether produced by X-rays or just by accident, generate variation, which is then acted on by selection. Sex, based on the same phenomenon of recombination that was crucial in validating the chromosomal theory of heredity, generates far more variation than is possible with random mutations alone, providing natural selection with an even broader palette of variation with which to work. The evolution of sex a couple of billion years ago, in some simple unicellular creature, stoked the fires of evolution and opened the way to the appearance of complex, multicellular life.
However, to say that variation is necessary for evolution, and that more variation made evolution faster, is in itself no argument for the prior existence of variation or an explanation for its nature. To view variation simply as the handmaiden of evolution is to leave a number of questions unanswered. What explains the range of what we think of as ‘normal’ variation? Why isn’t the range of normal variation greater than it is, or less? Indeed, why do creatures vary at all? Why do monsters seem to violate our innate sense of what constitutes the normal range of variation? Does this sense have any basis in fact? The emerging network view of genetic regulation allows me to propose – albeit tentatively – a reason why variation exists, and why it is particularly important for complex, multicellular creatures such as ourselves.
Monstrosities such as flies with extra wings and mice without heads might be explained in terms of disturbances to the connectedness – the topology, if you like – of genetic networks that are otherwise robust and modular. But variation – that is, what we think of as the normal range of variation expected in otherwise fully functional, healthy organisms – might also be regarded as a consequence of networks that allow a very large degree of latitude in the efficiency of their individual components. As the Seattle group showed, a network can take components of almost any quality or ability and channel their interactions to produce more or less the same product, provided the network is wired up in the correct way. Given the latitude allowed for the input values, it is to be expected that the output of a robust network – in this case, living organisms – might also be varied, although to a lesser degree, and at least in respect of characteristics that would not harm the overall function of the network – in other words, characteristics that correspond to normal variation.
Variation – the kind of variation without which natural selection cannot work – might, therefore, be simply a by-product of the particular way in which the genome directs the formation of each new organism – that is, as a robust network of interacting genes, each of which can vary hugely in its own properties and propensities. But this begs an even deeper question: why is it that the genome is organized in just this way, to produce organisms that are variously flawed, when it could have created organisms that are all identical, and all perfect – the nature-philosophical ideal, rather than the messy reality? The reason, I think, illustrates the pervasive power of natural selection. What is at issue here is not so much the degree of variation we see in natural populations, but the very existence of that variation – and this existence is itself subject to natural selection.
Genomes first started to create multicellular creatures more than a billion years ago. Until that time, the world was populated by unicellular creatures. When a single-celled creature divides into two, it produces two complete, adult creatures more or less instantly, with no need to create complex form from a formless egg. Unicellular creatures, therefore, do not really have ‘development’.15 The evolution of multicellular creatures, therefore, raised a novel requirement: the need to direct matters to ensure the safe passage between egg and adult; and to do this efficiently, and with the best chance of achieving a viable result, rather than a monster or nothing at all.
In evolution, all that matters is that an organism develops from a fertilized egg until it reaches adulthood, so that it can then reproduce and pass on its genomic heritage to the next generation. However, in all but the very simplest multicellular creatures, the course of development is so intricate that a single slip at any stage has the potential to threaten the integrity and viability of the whole. Given the uncertainties of the real world, any organism that develops according to an overly strict, step-by-step procedure has very little chance of reaching term, let alone sexual maturity. For every single, perfect organism produced in this way, large numbers of embryos might have started on their uncertain courses, just so that one might emerge as an adult without suffering the fatal consequences of some mishap along the way – a mishap which, given the complexity of the overall process, and the dependence of each step on the one preceding, would be virtually guaranteed to happen. Against potentially terrible odds, any organism whose genome evolved a strategy of development that improved this dismal success rate, even by a tiny amount, would inherit the Earth.
That is indeed what happened – the future belonged to those creatures whose genomes were organized flexibly, so that their development could not be completely derailed by relatively minor disasters. These organisms would have been more successful at reproduction than those with less flexible genomes, and so would have been favoured by natural selection in Darwin’s canonical struggle for life. With the passing aeons, selection would have continued to favour the flexible, so that the most successful genomes were those whose activities were organized into decentralized, modular networks, able to tolerate an ever greater range of variation in the properties of their components – at the small cost of some residual variation between the individuals so produced. Identikit perfection, if it were ever achieved, no longer exists.
More than a billion years later, the consequences of this primordial selection for flexible, network-based genomes are all around us, in the birth of yet more human beings with each passing minute, each one minutely formed, created with a reliability and flexibility which almost defies belief. As a consequence of this very reliability and flexibility, each new baby is a unique individual with its own potential to be an athlete, an architect or an actuary; its own predisposition to disease and response to medicine; its own likes, dislikes, needs, hopes, wants and dreams.
I suggest that variation, for all that we take it for granted, and for all its subsequent utility in evolution, is a by-product of the truly amazing piece of network engineering that creates form from the formless. And yet, the existence of this variation might be the price that each of us pays for having been born at all.