Genetic drift

Why can a variant that helps nobody spread through a small population anyway?

Ask what makes a gene variant common and the ready answer is that it must be doing something for its carriers. Useful things spread; useless things do not. It is a tidy story and it is the one most of us were taught.

Yet biologists routinely find variants at high frequency — sometimes at 100 per cent — that confer no advantage whatever, and occasionally ones that are mildly harmful. Nobody chose them. Nothing favoured them. They are simply everywhere. If selection did not put them there, what did?

Start with something that has nothing to do with biology. Toss a fair coin ten times and you will not reliably get five heads; you might get seven. Toss it ten thousand times and the proportion will sit very close to a half. The bias in the coin has not changed between the two experiments. What changed is how much room chance had to move the answer.

Now notice that reproduction is a sampling process of exactly that kind. Whatever the allele frequencies are in one generation, the next generation is built from a finite draw of gametes out of that pool. Nobody is being favoured; the draw is fair. But a fair draw from a small pool overshoots and undershoots, and next generation's draw starts from wherever the last one landed. Ask yourself what a fair sampling error looks like when its output becomes the following round's input, and you have the whole mechanism.

The consequences follow from that one observation. Because the error compounds rather than cancelling, frequencies wander. And a random walk on the interval from zero to one has two edges it cannot leave: at zero the variant is gone, at one it is the only version left. Wander long enough and you hit one of them. Every neutral variant, given time and no new mutation, ends up lost or fixed — never quietly persisting at some middling value.

Which one? Here the arithmetic is unexpectedly clean. For a neutral allele, the chance of eventual fixation is just its current frequency. A brand-new mutation in a diploid population of N individuals sits at one copy out of 2N, so its chance of taking over is 1/(2N) — small, but not zero, and nothing about that number refers to whether it does any good.

Now bring in size. In a population of a million, one bad-luck generation shifts a frequency by a hair; the walk is so slow that even feeble selection has time to steer it. In a population of thirty, a frequency can jump by a tenth in a single generation for no reason at all. The rough threshold that population geneticists use makes this explicit: selection only reliably wins when the selection coefficient is larger than about 1/(2Ne). Below that line, a variant is effectively neutral no matter what it does, because the noise is louder than the signal. That comparison — the strength of the push against the size of the random jitter — is the whole of Tomoko Ohta's nearly neutral theory.

One refinement matters in practice. The N that governs drift is not the number of animals you can count. The effective population size Ne is usually smaller, sometimes far smaller, because breeding is skewed: a lopsided sex ratio, a few males monopolising matings, or a past crash all cut it. Since Ne over time behaves roughly like a harmonic mean, a single bottleneck generation dominates the average for a long while afterwards.

That is why this stops being an abstraction in conservation work. Northern elephant seals were hunted down to a remnant of perhaps a hundred animals or fewer in the 1890s; they have recovered to hundreds of thousands, yet they carry strikingly little genetic variation, because the alleles that were lost in the crash were never coming back. Founder populations show the same signature in reverse — a rare disease allele that happened to be in one of the founders can end up at frequencies no selective story would explain.

The commonest misreading is that drift is a weak force that occasionally nudges things. It is better understood as the default, with selection as the thing that has to be strong enough to overcome it. Selection does not switch on above some population size; drift merely gets quieter, so ever fainter selective pressures become audible.

A second correction: drift is not the same as a bottleneck. A bottleneck is one violent instance of it. Drift is happening continuously, in every population, including large ones — just slowly. And it is not a source of variation at all. Mutation supplies the raw material; drift only shuffles and eliminates. Left alone, drift removes heterozygosity at roughly 1/(2Ne) per generation, which is why "small population" and "genetically depleted" tend to arrive together.

Finally, drift and selection are not rival explanations to be chosen between. They act at the same time on the same alleles, and which one dominates a particular variant depends on that comparison between the size of its effect and 1/(2Ne). A variant can be genuinely advantageous and still be lost by bad luck — most beneficial mutations are.

• How coalescent theory runs the same process backwards, reconstructing when living lineages last shared an ancestor.
• Why the neutral theory of molecular evolution predicted that most substitutions in DNA would be invisible to selection, and how that debate stands now.
• How conservation genetics estimates Ne from real pedigrees and marker data, and why the number so often disappoints.

Every term the collection defines is gathered in the glossary.