Evolution and Arithmetic

Chapter 16: Evolutionary Systems

← Two Systems, Hiding in One BacteriumBuilding Blocks, All the Way Down →

One Program, Modified

Step back from bacteria entirely and up to a much larger scale of problem: not how a single gene gets switched on, but how an entire new body part gets built. Take the simplest possible version of the question. An animal already has a leg. How, evolutionarily, would it ever get an arm?

The tempting answer is: reuse the leg's own program, with modifications. A limb is a limb — bone, muscle, blood vessels, nerves, roughly the same kind of internal engineering either way. Rather than evolving an entirely separate blueprint for an arm from nothing, build it out of the same construction toolkit already proven to work for the leg, and simply adjust a few settings: different proportions, different digits at the end, different final shape.

This turns out not to be a hopeful guess. It's well-studied developmental biology, and it comes with layers that map almost exactly onto what a sensible engineer would design if asked to solve this problem on purpose.

At the base sits a shared, generic construction program. The genes that grow a limb outward from the body and lay down its bone and muscle are, for the most part, the same whether the limb is on its way to becoming an arm or a leg. The same set of genes has been found at work in limbs as different as a horse's single-toed leg, a bat's wing and a dolphin's flipper. The deep machinery doesn't know or care which one it's building.

Above that sits a small set of identity switches, deciding which overall structure gets built at all. The best established is a gene called Pitx1, switched on in the developing hindlimb but not the forelimb. The evidence for its role is about as direct as biology gets. In 1999 two Harvard researchers, Malcolm Logan and Clifford Tabin, switched Pitx1 on in the developing wing of a chick embryo. The wing began switching on other genes normally found only in legs, and grew up partly leg-like: the joint bent the way a leg bends, the digits became more toe-like, and the muscles were laid out as in a leg.

And above even that sits a layer of fine local rule-making — the equivalent of a specialist called in to handle the shape, length and proportion of each part once the broad decision has been made. Different Hox genes, the same family that sets out the segments of the body, are deployed region by region within the limb to handle this kind of local patterning, distinguishing the upper arm from the forearm from the hand.

It's only fair to say that this tidy three-layer picture is a simplification of something messier. Two other genes, Tbx5 in the forelimb and Tbx4 in the hindlimb, were long thought to be the main arm-or-leg switches, and early experiments in chicks seemed to confirm it. But a more rigorous study in mice, published in 2005 by Logan's own group, found otherwise. A mouse forelimb grown with the hindlimb gene in place of its own still came out as a forelimb. Those two genes, it turned out, mainly get limb growth started. The arm-or-leg decision depends on positional signals set earlier still, along the body, before the limb bud begins to grow — Pitx1 among them, and Hox genes. The real system isn't three clean layers stacked neatly on top of each other. It's layers within layers, each one apparently needing a further layer underneath it once anyone looks closely enough.

That messiness sharpens the question rather than softening it. This is an elegant, efficient way to build biological variety — a shared core, reused everywhere, modified by switches, refined by ever more local rules. It is also, on its own evidence, a system with internal architecture, doing something considerably more sophisticated than "a mutation happened to help." And it depends on modularity: on genes whose effects stay contained within their own layer and their own limb, instead of leaking out and scrambling everything else at the same time. A gene that does many jobs at once — biologists call it pleiotropic — is hard to change safely, because a change that improves one job usually damages another. A change to a generic tool gene that weren't properly walled off from the switches above it would affect every limb and every layer at once, rather than improving one cleanly. So the question is how such a walled-off arrangement could have arrived gradually, one accidental improvement at a time, when the walls are what make gradual improvement safe in the first place.

Biologists are aware of the problem, and have tried to show how modularity itself could evolve. Most of that work is done in computer simulations, in which networks of switches evolve under selection. Two findings stand out. When the goals a network is selected for keep changing, but are built from the same sub-goals in different combinations, modular networks tend to emerge. And when every connection in a network carries a cost, as wiring does in a brain, networks also tend to become more modular. Other simulations are less encouraging. In one, modularity emerged only under particular conditions, and remained a rare variant even then; in another, gradual change never reached the modular arrangement at all, though copying existing parts of the network could. Whether modularity reflects some general principle of evolution, or arises only in special circumstances, is still described in the field as an open question.

Notice what even the favorable results take for granted. A simulation of evolving switches starts with a working network, a working way of copying it with variation, and a working test that rewards some versions over others. It shows, at best, how an organized system could become more organized. It doesn't show how the first organized system arose, or that the special conditions the simulations need actually held for real animals at the right time.

That is the shape this book keeps finding, wherever it looks closely enough. Shape, in Chapter 13, needed a system, in place, before the body-building problem could even be attempted. Instinct, in Chapter 14, needed a system, in place, before behavior could be inherited at all. The brain, in Chapter 15, needed a system, in place, before any of that inherited behavior could be stored or carried out. And now modularity itself — the very thing offered as the efficient shortcut that makes evolution's job easier — turns out to need the same thing: a working system, organized the right way, before it can do any of the efficient work being credited to it. Chance does not get handed a blank page and all the time in the world. Every time this book looks underneath one of evolution's more impressive tricks, it finds another trick, fully assembled, quietly making the first one possible.

Last revised 2026-11-19

← Two Systems, Hiding in One BacteriumBuilding Blocks, All the Way Down →