Evolution and Arithmetic

Chapter 16: Evolutionary Systems

← One Program, ModifiedWhere This Leaves the Toolkit →

Building Blocks, All the Way Down

The same appealing idea shows up again at a much smaller scale, and it deserves to be taken just as seriously here as it was for limbs. Some proteins really are built from reusable parts. Many proteins are made of distinct units called domains, each typically a hundred to two hundred and fifty amino acids long, each folding into its own compact shape and often doing its own job. And a process called exon shuffling can move a domain from one gene into another, producing a protein that never existed before out of parts that already worked elsewhere. If that were a general-purpose construction method, it would make building a new protein enormously easier than writing one letter by letter. So it's fair to ask what such a method would have to include, and what it actually does include.

Start with a distinction that is easy to get backward: the difference between a protein's scaffolding and its working core. It's tempting to assume the scaffolding is the cosmetic part — structural filler, safe to swap around — while all the real constraint lives in the working site itself. That assumption doesn't hold up. Scaffolding does mechanical work. In many enzymes, a hinged part of the protein swings shut over the molecule being worked on, and swings open again to release it, over and over, opening and closing the very pocket where the chemistry happens. Researchers studying one such enzyme, adenylate kinase, found that the opening of this lid is the slowest step in the enzyme's whole cycle — the step that sets the pace for everything else. The swing of the scaffolding isn't a background formality. It is a precisely built moving part, and building it out of generic, interchangeable blocks is not obviously easier than building the working site itself.

The working core has its own resistance to substitution, already met in Chapter 7: even a comparatively small protein can have positions that tolerate no change at all, others that tolerate only a narrow range of alternatives, and others that are far more forgiving. Some position-by-position precision runs through a working protein wherever you look. Not just any amino acid will do, almost anywhere.

Now suppose an engineer set out to design a system for building proteins from a stock of parts. The list of requirements would be long. Standard parts that can be snapped together. A common connector, so that any part can join any other. A catalog of the right size. A way of finding the right part when it's needed. Assembly rules for which parts belong together. And some way of trimming and adjusting a part that almost fits.

Nature's version has almost none of this, and doesn't need it, because it isn't aiming at anything. It works like this. Genes in animals and plants come in pieces: short stretches that get used (exons), separated by long stretches that are cut out before a protein is made (introns). A domain often sits on an exon, or a block of exons, of its own. Because the introns around it are long and get cut out anyway, a break-and-rejoin anywhere inside them — an ordinary recombination error, or a jumping gene dragging a neighboring exon along with it — can carry the domain into another gene without any precise cutting. The standard parts are the domains, which fold on their own. The catalog is simply whatever the genome already contains. There's no retrieval system and no assembly logic: the pieces land wherever the accidents put them, and selection discards what doesn't work.

There is one connector rule, and it matters. The cell reads a gene's message three letters at a time, so a piece dropped into a new gene must not knock the reading out of step. That works only if the cuts on either side fall at the same point within the three-letter reading frame — in the jargon, if the introns on either side are of the same "phase." A domain that meets that condition can be moved into another gene, or copied within one, without scrambling everything after it. A domain that doesn't, can't. And the genome bears this out. In one study of human genes, of the domains that sit on a single exon, 83 percent had matching cuts on both sides, far more than chance would produce. That's the fingerprint of shuffling: the domains that could be moved were moved, and the others weren't.

Some proteins are clearly built this way. A favorite textbook example is tissue plasminogen activator, the clot-dissolving protein used as a drug for strokes and heart attacks, which is assembled from several kinds of domain found also in other, unrelated proteins.

So the shortcut is real, and it doesn't require a factory. But doing without a factory has a price, and the price is the point.

First, it only works where the connectors happen to match. Domains whose surrounding cuts don't line up are effectively locked in place.

Second, it is not a general method of building proteins. Exon shuffling played its major role in one branch of life — animals — and mainly in one kind of protein: those that let cells stick to and signal to one another, the proteins a many-celled body needs. It played little part in plants and other groups, and the leading researcher on the subject has found no evidence that it built the master switch proteins described earlier in this chapter. Bacteria, whose genes aren't split into pieces at all, don't do it this way.

Third, it finds few winners. When researchers mapped which domain families occur side by side in the proteins of sequenced genomes, most families turned up with only one or two partners; a handful were very versatile. Across all of life, only a small fraction of the possible combinations is actually used. How many shuffles it takes to find one combination that works can't be counted honestly from the data available, and this book won't invent a number for it. But the sparseness itself says something: most combinations, it seems, either never worked or were never found.

Fourth, the seams still have to fit. Two domains joined end to end don't automatically cooperate. Where they meet, the protein usually needs adjusting — an amino acid here and there changed so the pieces sit together properly. And that adjustment is done by the same slow, letter-by-letter point mutation whose arithmetic Chapter 8 has already tested.

And fifth, the most important: shuffling moves domains; it doesn't make them. Every domain it carries had to exist first, folded and working, and most domain families are very old — many of the parts used in animal proteins were already present in the single-celled ancestors of animals. The modular shortcut, examined closely, doesn't make the hard problem go away. It moves it: from building a protein to building the parts the protein is assembled from. And that brings the question back, once again, to the origin of the parts — the one place none of the toolkit's tools has been shown to reach.

Last revised 2026-11-19

← One Program, ModifiedWhere This Leaves the Toolkit →