Antibody affinity maturation

Why does the body deliberately scramble the genes of its own immune cells at a rate that would be lethal anywhere else in the tissue?

Every other tissue in your body treats mutation as a catastrophe to be prevented: proofreading polymerases, mismatch repair, checkpoints that would rather kill a cell than let a damaged genome divide. Then, in a lymph node a few days into an infection, an enzyme is switched on that deliberately writes point mutations into a B cell's antibody genes — at a rate on the order of a million times the background rate of the genome around it. That figure I am recalling rather than deriving, but the order of magnitude is not in doubt.

That is not a lapse in quality control. It is quality control run backwards, on purpose, in one place, for a few weeks. What could be worth it?

Start with the problem an antibody has to solve. Its binding site must fit a molecular shape no ancestor of yours ever met. The body's first answer — randomly cutting and rejoining gene segments so the starting repertoire covers almost any shape — gets you a cell that binds the pathogen somewhat. A lottery over an enormous space of shapes gives you near-misses, not winners.

So you hold a receptor that binds weakly and want one that binds tightly. Could you compute the improvement — work out which amino acid to change? Nothing in a cell can do that. Predicting how a substitution shifts binding energy is hard for us with the structure in hand; a lymph node has no structure and no calculator. What is left when you cannot compute an answer but can recognise one?

Vary and select. And that immediately tells you the two things the body must build: a way to generate variants of an existing receptor, and a test that scores them. Both exist, and they are separated in space.

The variation happens in a germinal centre, a temporary structure forming in a lymph node after activation. In its dark zone, B cells divide fast while an enzyme — activation-induced cytidine deaminase, AID — chemically alters cytosine bases in the antibody's variable region; error-prone repair of that damage leaves point mutations behind. Do the arithmetic on the rate. The heavy and light variable regions together run to roughly seven hundred bases; at about one mutation per thousand bases per division, that is around 0.7 mutations per cell per division. Almost exactly one. Not a scramble at all — a tuned drip. Faster and most daughters would carry a wrecked receptor; slower and a week of divisions would explore nothing.

Now the test, in the light zone. Antigen is held there on a network of follicular dendritic cells, and it is scarce. A B cell must grab some with its new receptor, internalise it, chop it up and display the fragments to a follicular helper T cell — and helper cells are scarcer still. Do you see how neatly that converts binding strength into a survival signal? A better receptor captures more antigen, displays more peptide, wins more help, and lives; a worse one gets no signal and dies. Nothing measures affinity; competition for two limited resources does the measuring.

Then the move that makes the whole thing powerful: survivors are not finished. Many return to the dark zone, divide, mutate again, and come back to be tested again — cyclic re-entry, demonstrated directly by labelling cells in one zone and watching where they go. A single round of one mutation per division can only take small steps; twenty rounds of small steps, each one filtered, climbs a hill no single jump could reach.

So affinity maturation is Darwinian evolution running inside one organ, over days rather than millennia, on a population of a few thousand cells. The end result is an antibody binding orders of magnitude more tightly than the one the response started with — I will not put a figure on it, because the improvement varies enormously with the antigen and how it is measured.

The falsification test. Everything above rests on one claim: the improvement comes from mutation-plus-selection, not from some directed refinement. Remove the mutator and the improvement should vanish while everything else continues. It does. People and mice lacking functional AID form germinal centres — unusually large ones — and mount responses, but their antibodies show no somatic mutation and no affinity gain. The refuting observation would be an AID-deficient animal whose affinity climbed anyway over three weeks: something else would be tuning the receptor, and the mutational account would be wrong.

Three refinements connect the steps.

First, the mutation is blind but the targeting is not. AID is aimed at the antibody locus by features of its transcription, so hypermutation is confined to a small region of one gene. That confinement is the safety mechanism — and it is imperfect: AID acts on off-target genes too, and that activity is implicated in the translocations behind several B-cell lymphomas. The danger in the opening question is bounded, not abolished.

Second, this is not the gene rearrangement that built the starting repertoire. That happens before a cell meets any antigen and creates diversity by shuffling whole segments; this happens after, and refines one receptor by single-base edits. The two are opposite moves: exploration first, then exploitation.

Third — the honest complication — randomly editing a receptor can create one that binds you. Tolerance was established before hypermutation existed in that cell, so the germinal centre can regenerate self-reactivity from scratch. There are checkpoints against it, and how they work is an active question rather than a settled one. Which is why the deletion of self-reactive lymphocytes in the thymus is a separate story: that mechanism screens a repertoire before it is used, this one screens a repertoire while it is being rewritten.

• Why pathogens such as HIV and influenza defeat maturation by changing faster than a germinal centre can climb.

Every term the collection defines is gathered in the glossary.