Evolution and Arithmetic

Chapter 8: Weasels

← It's All a GamePushback →

The Paradox of the Lock

There's one more matter worth raising, parts of which have already come up in passing, and it bears directly on any attempt to explain how evolution works — not just point mutation acting on junk DNA. It comes into focus when two things already discussed are considered together: the need for precision at specific positions, and the idea of a lock. Taken together, they create something close to a genuine paradox.

Chapter 7 described studies that tested real proteins position by position, to see how much flexibility exists at each spot. In one, a small, common protein module called a PDZ domain, 83 positions long, 20 positions stood out as highly sensitive: on average, swapping in other amino acids there did serious damage, while most of the other positions barely mattered. In the other, a much larger protein of 360 amino acids, 42 positions tolerated no substitute at all.

Suppose, for the sake of argument, that each of those critical positions needs one particular amino acid. That's a simplification — the first study measured how sensitive each position was, not an all-or-nothing rule — but it lets us follow where the arithmetic leads. Without any mechanism to lock a correct amino acid in place once it's found, a blind search for either of these proteins is effectively hopeless. Finding the 20 correct amino acids at the 20 critical positions in the smaller one, with twenty possible amino acids at each spot, means one right answer among twenty multiplied by itself twenty times — on the order of 1026 possibilities. For the 42 critical positions in the larger protein, it's one among roughly 4.4 × 1054. Adding twenty-two more such positions doesn't nudge the number up modestly. It multiplies it by about 1028 — a number already far larger than the 1026 we started with. (The arithmetic: Twenty and Forty-Two Positions.)

There's one way, in principle, to bring a search like this down to something manageable: lock each correct amino acid in place the moment it's found. This is the same trick the Weasel program relies on — find one correct letter, lock it, move on to the next, and the remaining search shrinks dramatically at every step. The obvious question is whether anything like this is possible in the real world, and that question goes straight to the heart of evolutionary theory's central mechanism: natural selection. If a change is beneficial, the reasoning goes, it will be selected for, and the gene carrying it will be preserved — effectively locked in place, because the organism now depends on it. Once that first change is locked in, a second can occur on top of it, and the whole process can reach a working result in a manageable number of steps — subject to everything else this chapter has raised, but nowhere near as long as an unlocked search would take.

The problem is junk DNA. This chapter is working through whether junk DNA can be the source of new, functional genes — other possibilities will get their turn later — so for now, the raw material on the table is junk, and only junk.

So ask the obvious question: if a working protein needs, say, twenty specific positions to be exactly right, does hitting just one of them correctly, somewhere in an ocean of otherwise meaningless DNA, do anything at all? Start with a stretch of junk — by definition, doing nothing useful. It mutates repeatedly, mostly in places where precision doesn't matter. Then, eventually, purely by chance, a mutation lands on one of the twenty positions that does matter. Does that moment suddenly make the whole stretch beneficial enough to be selected, and therefore locked? Almost certainly not. One correct position out of twenty isn't remotely enough to produce a working protein — there's nothing yet for selection to notice or preserve. So no lock happens. The search simply continues, unprotected, the same as before. The real question, then, is this: how many of the needed positions have to be correctly in place before the whole thing becomes functional enough to finally be worth locking?

Here's where the numbers matter. The simulation earlier in this chapter locked each position the instant it was correctly hit. This time, the same simulation was run for twenty precise positions, with twenty possible amino acids at each, but with the lock deliberately delayed: the simulation waits until some chosen number of the twenty positions happen to be correct at the same time before locking any of them, then locks the rest one at a time as they occur.

The results make the shape of the problem clear. If the lock is allowed to happen after just one, two, or three positions are correct, the whole twenty-position task finishes after somewhere between about 1,200 and 1,400 random draws — trivially fast. Push the threshold further, and the cost starts climbing, gently at first: lock after four correct, and it takes about 1,975 draws; after five, about 3,248; after six, about 8,955. Past six, the climb turns sharp. Lock after seven, and the number jumps to roughly 107,271 draws. Lock after eight, and it's roughly 520,459. Lock after nine, and the average balloons to around twenty million — if the run finishes at all. (All the results: The Lock Simulations.)

In other words: a protein that needs twenty specific positions to be exactly right will never be found in any realistic amount of time unless something makes it functional enough to lock in the positions already found well before six or seven of the twenty are in place. A lock arriving at seven or eight already makes the remaining search inordinately long. Any later than that, and completing the search becomes, for all practical purposes, impossible.

Now try the same experiment with forty-two precise positions instead of twenty.

With more positions to find, the numbers behave in much the same way, in one important respect: there's still a threshold below which the search stays entirely reasonable. If the lock occurs once anywhere from one to five of the forty-two positions are correct, the whole task completes in roughly three to just under four thousand draws — well within reason, depending on the organism's rate of reproduction and the other factors, like spread through a population, discussed earlier in this chapter.

But then the numbers begin climbing, the same way they did with twenty positions. Lock after six correct, and the total rises to 4,652. Seven brings it to 5,796. Eight brings it to 10,379. Past that point, the climb turns much steeper: 58,383 if the lock waits for nine, 145,983 for ten, and 989,034 for eleven. By twelve, the results scatter widely — sometimes failing to finish even after twenty million draws, sometimes finishing in the low millions, and occasionally, when unusually lucky, in under a million.

There's a pattern worth pulling out of these runs. When forty-two positions need to be precise, the point where the search becomes effectively impossible arrives at around 30 percent of the total — a lock at twelve or thirteen. When there are only twenty, the same breakdown arrives at close to half the total — a lock at nine or beyond. Run the same experiment with a hundred precise positions, and the threshold falls further still: wait for twenty-five, a quarter of the total, and the search already needs upward of twenty million draws. The pattern is consistent, and it runs in only one direction: the more precise positions a protein needs, the smaller the share of them that can be in place before the search becomes impractical.

One honest qualification belongs here. These simulations follow a single line of descent, one change at a time. Bacteria don't search that way; they search in parallel, in unimaginable numbers. Chapter 3 counted every bacterium that has ever lived, each trying a fresh chain every twenty minutes for three and a half billion years, and found about forty-five drums' worth of tries. Twenty exact positions, on their own, make a twenty-seven-drum target. If those twenty positions were all that mattered, with the rest of the chain free to be anything, bacteria searching blindly and in parallel could have hit them many times over. Forty-two exact positions are another matter: a fifty-five-drum target, ten drums beyond everything bacteria could ever have tried. And for animals like the hominids, with populations in the hundreds of thousands and generations measured in years, nothing like that parallel search exists. So the paradox bites hardest exactly where it matters most for this book: proteins with many precise positions, and creatures that reproduce slowly. (The arithmetic: Twenty and Forty-Two Positions.)

To my mind, this creates something close to a genuine paradox. Start with junk DNA, and any protein requiring more than a handful of precise positions can never be reached by a step-by-step process — one amino acid at a time — unless that junk DNA becomes beneficial well before all the needed positions are actually in place. Is that impossible? Not necessarily; it may turn out to be true in some particular cases. But unlikely, to my mind, yes. Gaining real function while only partly precise seems to run against everything else already established about these proteins. If changing even one of twenty critical amino acids does serious damage to the protein, how could that same protein have already been useful back when only half, or fewer, of those positions were correctly in place?

And this calculation, demanding as it already is, only accounts for the positions that need absolute precision. The studies that examined proteins position by position found something more nuanced than a simple two-way split. Some positions really were entirely flexible — almost any amino acid would do. But many others fell somewhere in between: not perfectly rigid, but not freely open either, tolerating only a limited set of substitutions rather than all twenty. Factor in these partly constrained positions as well, and the numbers only get worse — making it that much harder to believe any slow, step-by-step search could ever fill every one of these slots correctly.

There's another study worth bringing in here, because it adds yet another layer to the same problem. Researchers took a human DNA-repair enzyme and made large collections of mutants, each carrying several random amino acid changes at once — on average about two in one collection, about four and a half in another, and about six in a third — the way a real history of accumulating mutation would. From how many mutants still worked, they calculated the chance that a single random change knocks the protein out: about one in three. And the effects multiply. With four or five random changes, most of the mutants were dead.

Bring that finding into the picture just built, and the difficulty compounds rather than simply adds. A protein that is already complete and fully working is knocked out one time in three by a single random change, and most of the time by four or five of them — and most of those changes land at positions that supposedly don't matter much. If that is what a few unimportant changes can do, how could a protein with half or more of its critical positions still wrong have been working well enough to be kept?

Some further problems with the making of new proteins will come up at the end of a later chapter, on how a protein is made.

Last revised 2026-11-19

← It's All a GamePushback →