Chapter 7: Junk DNA and a Demolition Tool
The Junk That Wasn't
Once the structure of DNA was worked out in the 1950s, and scientists understood how its information gets turned into proteins, an obvious question followed: how much of the DNA is actually doing that job, spelling out proteins? The answer, when it came, was surprising. Only a small fraction: in humans, under two percent. Other findings pointed the same way. The amount of DNA an organism carries turned out to bear little relation to how complex it is. An onion has about five times as much as a person, and some salamanders have nearly forty times as much. Much of the extra turned out to be repetitive: the same stretches copied over and over, many of them recognizable as broken copies of genes, or as the remains of "jumping genes" and ancient viruses that had copied themselves into the genome long ago. The name for the leftover DNA had been floating around for some years, but it was the geneticist Susumu Ohno who made it stick, in a 1972 paper: junk.
Alongside that came a looser idea, common in popular accounts if not among specialists: that all of life could in principle be built from nothing more than the information that makes proteins. I recall reading, years ago, the claim that as long as the DNA specified which proteins to make, that alone was enough to account for something as staggeringly complex as a person. I remember thinking, even then, that this was one of the more remarkable claims I'd encountered in biology, and not in a good way. Specialists knew better. In 1961 two French researchers, François Jacob and Jacques Monod, showed that genes are controlled by switches — stretches of DNA that make no protein themselves but decide when a gene gets used — work that later won them a share of a Nobel Prize. How much of the genome such switches took up, though, was anyone's guess.
A person isn't just a bag of proteins. There's shape. There's organization — where everything goes, and in what arrangement. There's instinct. There are countless minute details that have to be specified somehow, somewhere. The idea that a blueprint for the raw components alone, with no further instructions, could account for all of that never sat right, long before there was any direct evidence one way or the other.
The junk idea rested on evidence. But it also suited the leading theory of the time very well, and that helped it settle in. Evolutionary change was understood to happen mainly through very small edits — one letter of DNA at a time. That picture runs into an immediate problem if every stretch of DNA is doing something important. Tinker with DNA that codes for a protein the organism needs, and you don't get a smooth improvement; you get a broken protein, with no guarantee anything better arrives before the damage catches up with the organism. A genome that's mostly inactive filler eases this. There's room to tinker without consequence, room for slow trial and error to work at its own pace. Ohno himself thought new genes came mainly from duplicated copies of old ones, not from junk. But junk DNA gave the theory a comfortable safety margin, and it was welcomed partly for that reason.
Fast-forward to the twenty-first century, and a new set of tools. A large, well-funded research effort called ENCODE set out to map systematically what the non-coding parts of the genome were doing, if anything. The results, published in 2012, seemed to upend the old picture: the project reported that around eighty percent of the human genome showed some kind of biochemical activity — far more than the small coding fraction. Newspapers ran with it. The idea of "junk DNA" was, by several accounts, given up for dead.
The pushback from parts of the scientific community was immediate. It wasn't one kind of objection — it was four different things, tangled together.
One was a numbers-based argument. Every generation, copying mistakes happen across the genome at a roughly known rate. If a large share of the genome is doing something important, a large share of those mistakes will be damaging ones. And a population can only absorb so many damaging mistakes per generation before more harm piles up than the ordinary weeding-out of survival and reproduction can clear away. Run the numbers on this, using reasonable assumptions about mutation rates, and you get a ceiling: something in the rough neighborhood of ten to twenty-five percent of the genome, at most, could be doing important work without the population collapsing under its own copying errors. One of the most prominent scientists making this argument put the stakes bluntly: if the genome really is mostly functional, then the standard picture of slow, gradual, undirected evolutionary change can't account for the human genome as we find it. Other researchers have pushed back on how tight that ceiling is, and that part of the dispute continues.
A second objection had nothing to do with mutation rates. It was about what ENCODE meant by "functional" in the first place. To count a stretch of DNA as biochemically active, ENCODE's tests only needed to see something happen there — a protein briefly attaching to it, or the stretch getting copied into a scrap of RNA even once, in any of the cell types tested. Critics pointed out that this is a remarkably low bar. A protein can bump into and briefly stick to almost any stretch of DNA without that contact meaning anything; a segment can get copied into RNA by accident, with the resulting RNA doing nothing and going nowhere. Something happening near a stretch of DNA is not the same as that stretch doing a job the organism depends on. Critics suggested that entirely random, computer-generated DNA, run through the same tests, would likely register plenty of the same "activity" — which would mean the tests were picking up background noise as much as function. So the second objection wasn't about whether the math could support so much of the genome being useful. It was that ENCODE's definition of "useful" was so loose that the eighty percent figure may never have meant what the headlines took it to mean.
A third line of evidence came from comparing species. If a stretch of DNA does a job the organism depends on, most changes to it will do damage, and the animals carrying those changes will tend to leave fewer offspring. Over millions of years, such a stretch should change much more slowly than DNA that does nothing. Comparing the human genome with those of other mammals shows exactly this pattern in some places and not in others, and the share that has been held steady in this way comes to roughly eight to fifteen percent of the genome, depending on the method; one widely cited study, from Oxford in 2014, put it at 8.2 percent. That estimate has uncertainties of its own — a stretch can matter without being held steady across species, for instance if it took on its job only recently — but it doesn't depend on mutation rates, or on anyone's definition of "function." That same year, many of ENCODE's own leaders joined in a paper acknowledging that biochemical activity and evolutionary function are not the same thing, and that the share of the genome showing activity is far larger than the share that has been held steady.
But there was a fourth thread running alongside the technical arguments: a sentiment, expressed by more than a few voices at the time, that came closer to "this can't be true, because if it were, there wouldn't be room left for evolution to work" — treating the conclusion as something to be ruled out because of what it would cost the existing theory, rather than something to be weighed on the evidence. That's a different kind of objection from the mutation-counting argument, and a much weaker one. A theory needing a certain finding to be true is not, by itself, evidence that the finding is true. The original junk DNA idea was welcomed partly because it was convenient for the theory, and some of the resistance to overturning it carried that same convenience with it.
Whatever the final tally turns out to be — and it isn't settled — two things are now clear. The first is that DNA does a great deal more than spell out proteins. Much of the DNA that matters isn't protein-coding at all: stretches that work as switches, turning genes on or off, up or down, in different tissues and at different times, without changing the genes themselves. And the instructions an organism carries go far beyond a parts list. Somewhere in the DNA are the instructions for shape — how a single cell knows to become the front end of an animal rather than the back, where a limb goes, how many fingers to build. Somewhere are the instructions for instinct — behaviors an animal never has to learn. Somewhere are the instructions for building a brain, whose trillions of connections make thought, memory, and perception possible. None of that shows up on a list of proteins, however complete. Each of these is a large enough subject to deserve its own chapter later in this book.
The second is that a large reserve of DNA probably still does nothing in particular. On most estimates, the DNA that does useful work — protein-coding genes, switches, and everything else — adds up to well under half the genome, and perhaps as little as a tenth. Much of the rest is the debris of copying: broken genes, and the decayed remains of jumping genes and ancient viruses.
So here is where this leaves us. DNA is not simply a parts list for proteins, with everything else as filler; the protein-only picture is gone. But the junk hasn't disappeared. There is still a large stretch of genome, free to change without harm, for the Standard Model to work in — and what it can build there is a question for the next chapter.
Last revised 2026-11-19