Original antigenic sin

Why does a person meeting this year's flu keep making antibodies shaped for the strain they first met as a child?

Immune memory is usually told as an unambiguously good story: meet a pathogen, keep a record, respond faster next time. A companion piece in this collection asks why that works, and notes that influenza defeats it by changing its surface faster than memory decays — the memory stays intact, but the face no longer matches.

The odd thing is not that the old memory fails. It is that the old memory wins anyway. Draw blood from someone just recovered from this winter's influenza and the highest antibody titres are often not against this winter's strain at all, but against the strain circulating when they were small. Thomas Francis Jr. observed this in serological surveys and named it, around 1960, original antigenic sin — a name misleading in two directions at once.

So why should a body meeting a new thing answer with an old answer? Not merely fail to update, but actively produce the wrong-shaped response?

Begin with what makes memory fast, because the whole phenomenon falls out of it.

A naive response has to solve a search problem. Among an enormous library of B cells, each carrying a receptor for one essentially arbitrary shape, the few that fit must be found, activated, and put through the germinal centre — rounds of mutation and selection that refine their binding. That takes something like one to two weeks. Memory has already done all of it: the surviving clones are numerous, their receptors refined, their activation threshold lower. They can be secreting antibody in a few days. (Those timings are recalled approximations, not measured values for any particular infection.)

Now ask what a drifted influenza strain presents. Not a wholly new surface — a partly changed one. Some regions of the haemagglutinin protein have mutated away from what the immune system saw before; others, constrained by the job the protein does, are essentially unchanged.

Put those two facts together and the outcome is forced. The drifted virus arrives carrying both old and new features. Memory cells recognise the old ones immediately and go off within days. The naive cells that would recognise the new ones are exactly where they were in an ordinary primary response — rare, unrefined, needing a week or two. It is a race with a head start of roughly a week, and there is one plausible winner.

But losing a race is not the same as being suppressed, and here the argument needs its sharpest step. Why do the naive cells not simply arrive late and contribute anyway? The proposed mechanism is epitope masking: the antibody the memory cells pour out binds the virus at the shared sites, which both clears antigen from circulation and physically covers the particle. A naive B cell needs to encounter intact, unbound antigen displaying its target in order to be activated at all. The fast response therefore removes the very stimulus the slow response depends on. This has been argued theoretically and modelled explicitly by Zarnitsyna, Antia and colleagues, and is supported experimentally — but I should be clear that it is the leading explanation rather than a closed case.

So the response to strain B is dominated by antibodies fitted to strain A, aimed at whatever A and B have in common. Not a memory failure — a memory success, occurring at the wrong target.

Is this testable? Very cleanly, because it predicts something no other model does. If your response is imprinted by your first exposure, then susceptibility to a genuinely novel influenza should track the year you were born — which strain family was circulating when you were a child — rather than your age. Gostic and colleagues tested exactly that against the human cases of avian H5N1 and H7N9. The severe cases sorted by birth cohort in opposite directions for the two viruses, each sparing the cohort whose childhood strain shared its haemagglutinin group. The observation that would have refuted it: severity rising or falling smoothly with age, with no birth-year signature. That is not what the data showed.

Notice what that does to the word "sin": imprinting there was protective, giving lifelong cross-group cover against a virus nobody had met.

Three refinements, each correcting a natural over-reading.

The name is wrong twice. "Original" overstates it: the pattern is better described as seniority — earlier exposures outrank later ones in a graded way, rather than the first alone holding a monopoly. And "sin" imports a judgement the biology does not support: the mechanism that biases you toward last century's strain is what buys you a defence in days instead of weeks, and can be what saves you from a virus you have never met.

Nor is it a hard ceiling. Vaccination and infection do raise antibody against genuinely new sites, and they also raise titres against strains met long ago — "back-boosting", the same imprinting seen from the other side. The response is biased, not frozen, and how far depends on antigenic distance: push far enough from anything in the repertoire and a proper de novo response is mounted, because nothing is conserved enough for memory to grab first.

Finally, the causal core is a race, not a preference. Nothing decides to prefer the old shape; speed alone produces the bias, because whichever arm finishes first consumes the antigen both arms needed.

• Why vaccines aimed at the conserved stalk of haemagglutinin are attractive, and how imprinting complicates evaluating them.
• Whether the same imprinting logic applies to coronaviruses and to dengue, where a second infection can be more dangerous than the first.

Every term the collection defines is gathered in the glossary.