As biologists figure out more about how life is, they can then figure out how it got to be that way. First there were genes. Mendel noticed that somehow the wrinkles on wrinkled peas could be transmitted down through the generations, even if some of those generations had no wrinkles at all. It turned out that the wrinkles were the result of a gene; a different version of the gene produced smooth peas. For much of the twentieth century, evolutionary biologists worked out how changes in genes produced evolutionary change. A mutation that alters one position in a gene (or chops out a whole chunk of it) can alter the protein it encodes. As the proteins on a virus mutate, for example, their shape becomes harder for an immune cell to recognize.

But towards the end of the twentieth century, it became clear that the protein-coding sequence is not the whole story. For example, many genes are equipped with on-off switches. Only if other proteins toggle these switches on will a gene produce its own protein in a particular place and time. A slight tweak to one of these switches can produce a drastic change--adding or subtracting legs from a segment on an insect's body for example. Other proteins destroy other proteins, while others enhance their supply. Some genes create proteins that can only work when they fuse to proteins made by a different gene. You can think of the genes as pieces of a complicated circuit, evolutionarily wired for some particular job, such as sensing a molecule or telling time.

How then do networks evolve? At first this can seem like an insurmountable problem. Consider a network of three genes that can only do a job if all three genes are working together. How then could the network evolve from two genes, let alone one? This is the basic "irreducible complexity" argument you sometimes hear from the Intelligent Design camp. They'd like you (or at least your local board of education) to think that you can't get there from here, and that someone must have designed the network from scratch. In reality, many scientists are now probing genomes to figure out how networks evolve, generating detailed hypotheses, testing them, and publishing their results--yet never once finding the need to utter the phrase Intelligent Design.

The key to network evolution lies in yet another way genes can mutate. Instead of just a small segment of its DNA changing, it's possible for an entire gene to get duplicated. Gene duplication happens a lot, judging from the many families of similar genes both in our own genome and those of other species. A copied gene would initially play the same role in the original network. But as it gradually mutates, it can take on a new function. Can it take up a new role in a new network? One clue that the answer is yes is that many networks are made up of related genes. Some researchers have proposed that all the genes in a network (perhaps even an entire genome) have to get duplicated at once in order to create a new network. But this large-scale copying may come with its own trouble: somehow, all the copied genes would have to stop interacting with the old network.

In the current issue of EMBO Reports, scientists at the University of Manchester in Great Britain offer a more humble way to build a new network. They suggest that it can happen one duplicated gene at a time. Imagine that one gene in a three-gene network gets duplicated. A mutation prevents it from interacting with the original three. Then it gets duplicated in turn, and these two genes start interacting in a tiny network of their own. Another duplication, and there are three genes at work in a fully-functional network that's completely isolated from its parent.

It would have been vaguely interesting if the scientists had stopped there, but then they figured out a way to test their hypothesis. They studied a family of genes that produce molecules called basic helix-loop-helix proteins (bHLH). These genes form several networks in our own bodies and in those of other animals. By linking with one another in different combinations, they can do all sorts of work in the cell, from sensing signals from the environment to keeping cell division under control. The history of these networks, the researchers realized, should be preserved in the genealogy of the genes. Say that some ancestral bHLH network was copied all at once. Then each gene of the new network should be most closely related to the gene playing the same part in the old network. But if, as the scientists propose, new networks are built a gene at a time, then all the genes in a new network should be closely related to each other, and only distantly to the old network. When they drew the bHLH family tree, that's what they found.

What's particularly remarkable about this work is what it means about the way new networks evolve. Each one budded off from an old network as a single duplicated gene. But over time, as the new network expanded with additional gene duplications, the new network came to look and act a lot like the old one. Each network, for example, is organized around a hub of a few genes that can interact with a constellation of other genes. Stephen Jay Gould famously asked whether life would take the same form it has today if you replayed the tape. Gould thought that there were so many contingencies that could push life off on another path that the answer must be no. But when it comes to gene networks, it appears that the tape may play just about the same.

(Update, 3/1/04 8 am: Link to paper fixed, along with a few typos.)