When a building goes up down the street, there is a drawing on site: how high this wall runs, where the window goes, how the pipes are routed. It is hard not to assume a genome is that kind of drawing — every part of a body with its own corresponding line on it.
But a fertilised egg contains no shape whatsoever. It holds a set of instructions executed in time: first decide which end is the head and which the tail, then cut the middle into segments, then let each segment work out what to become. A drawing describes the result; instructions describe only the moves.
In the late 1970s people began breaking those instructions on an industrial scale and looking at the monsters that came out. The monsters gave away the structure of the instructions.
Start with two famous monstrous flies. In one, a leg grows where an antenna belongs — not a stump, but a properly jointed leg sticking out of the head. In the other, a second pair of wings grows where a pair of halteres should be (a fly's hindwings are reduced to two small knobs that handle balance in flight), giving a four-winged fly.
Mutations of this kind all land on a small group of genes, later called Hox genes. What matters is how they work: they do not build a leg, and they do not build a wing. Each governs one stretch of the body, and switching it on amounts to telling that stretch "you are the second thoracic segment"; the stretch hears this and calls up the whole wing-building program by itself. So breaking one does not give you a fragment. An entire stretch of body takes on a different identity: the segment that should have made an antenna takes itself for the leg-making segment, and dutifully makes a leg.
This is more like a hotel front desk handing out key cards than like a drawing. The desk does not arrange the furniture; it only tells you that you are on the third floor, and the third floor knows what the third floor looks like. Hand someone the wrong card and they do not lose an arm — they simply walk into a different room.
Then comes the least reasonable part of it. These genes sit in a row along the chromosome, and the order they sit in corresponds one for one with the order of the regions they govern: the ones at this end run the head end, the ones at that end run the tail end. It holds in flies; it holds in mice and in us, where there are four such clusters, thirty-nine genes in all, likewise lined up — and in vertebrates even the timing follows the same order, the earlier ones opening first.
Why they queue up has no agreed answer. And it is not an iron law — in some animals the cluster has broken apart, the correspondence went with it, and their bodies develop perfectly well.
How was this list of instructions compiled? In 1980 Nüsslein-Volhard and Wieschaus used the dumbest method imaginable: generate random mutations in bulk, look at embryos one by one, pull out every one with the wrong number of segments or with scrambled structure inside a segment, and work backwards to the instruction that broke. Together with Lewis, who had studied the four-winged fly earlier, they shared a Nobel Prize in 1995. The list was reconstructed from what broke, not read off from what worked.
A fly's eye is a compound eye assembled from hundreds of small units; yours is a single eye with a lens. Two approaches this far apart, and you would probably guess the genes underneath are separate too.
In 1995 Gehring's lab forced a gene called eyeless to switch on in the legs, wings and antennae of fruit flies, and eye structures grew in those places. Better still, the corresponding mouse gene (named Pax6) put into a fly does the same job — and what grows is still a fly's compound eye, not a mouse's eye.
It is easy to hear this as "the genes for making eyes are shared", which gets it backwards. The mouse gene, inside a fly, starts up the fly's own eye-building program. So it is not a drawing of an eye, it is a switch; what the common ancestor left both sides is that switch and the set of parts hanging under it, not the eye. Structurally the two eyes were very likely built separately, while what they call on underneath is the same ancient kit. This situation is called deep homology.
The same story repeats for appendages. A gene in charge of "push something outward from here" switches on at the corresponding spot in an insect's leg, the margin of a butterfly's wing, the tube feet of a sea urchin and the limb bud of a vertebrate. Anatomically these protruding things have nothing to do with one another; there was no single ancestral limb they all descend from.
Why can such a toolkit sit unchanged for hundreds of millions of years? Because these genes sit at the very top of the program. Change one upstream switch and everything downstream that depends on it changes too, and in the overwhelming majority of cases the embryo does not survive to be born — a variant like that is never passed on even once. So the changes that do accumulate almost never land on the proteins these genes encode. Where do they land instead?
One caveat on how far the line can be pushed. Calling Pax6 "the master switch for eyes" is now regarded as overstatement: it is switched on in plenty of places that never make an eye, and it does not work alone — it sits in a small circle of genes that activate one another.
A giraffe has a nerve that leaves the brain to serve the throat. Brain and throat are less than a hand's width apart, but the nerve does not take that route: it runs down the whole neck into the chest, loops around a major vessel beside the heart, and climbs the whole neck again. Several metres of travel to cover less than a hand's width.
In a fish the same nerve runs straight. A fish has no neck, and that vessel sits near the gills, so the detour costs next to nothing. As necks lengthened generation by generation, the nerve's detour was stretched out along with them. No single generation was worse off — each was only a little longer than the last — and "cut it and reconnect it" is a move no generation could make.
Development has no step in which you tear everything down and start over. Each generation's body is modified from the previous generation's running program, and every step along the way has to be an embryo that survives. In 1977 the French biologist François Jacob gave it a phrase still in use: evolution is not an engineer but a tinkerer — a tinkerer designs no parts for the job, uses whatever is at hand, and turns out something that works but rarely looks respectable inside.
So where do the changes usually land? A gene tends to be used at several places in the body and at several times. Changing the protein it encodes changes every one of those uses at once, which is usually a catastrophe. But each individual use is generally controlled by its own stretch of regulatory sequence — think of these as a row of switches at the gene's door, each governing one occasion. Delete one and only that one light goes out.
The cleanest example is the threespine stickleback. Marine sticklebacks carry a pair of hard spines under the belly, and populations stranded in lakes across North America after the ice retreated have lost that pair over and over. On investigation the gene governing the structure is entirely intact, its jobs elsewhere in the body going on as usual; what was deleted is the switch at its door that works only at the pelvis. More telling still, many unconnected lakes deleted the same stretch independently — that sequence is itself prone to breaking. Put it back and the spines come back.
The stronger claim, that morphological evolution is mostly a matter of changing switches, is still contested; others argue that changes in the proteins themselves are systematically underrated. What holds is the weaker version: in the cases worked out so far, what changed was the switch.
Tinkerers' work has one more signature: it keeps pressing one thing into service for something else. The "thumb" a giant panda strips bamboo leaves with is not a finger but an enlarged sesamoid bone in the wrist — all five fingers were already committed. The piece on adaptation and constraint covered historical baggage; here is where the baggage comes from. Not a wrong choice made long ago, but the requirement that every step be completed on a living embryo.