Smell coding

How can a nose tell thousands of smells apart using only a few hundred kinds of receptor?

You can tell coffee from burnt toast from wet dog from petrol, and you can tell them apart from a molecule synthesised for the first time last year in a laboratory. Yet the nose is not improvising: it has a fixed, inherited parts list. Humans carry roughly 400 working odorant-receptor genes — a large family by the standards of the genome, but a very small number compared with the count of things that smell of something.

So the arithmetic does not work for the obvious model. If each receptor were a lock waiting for its one key, we could recognise 400 odours and be baffled by everything else. What has to be true instead?

Notice first what the lock-and-key model would predict. It would need a receptor waiting for a molecule that has never existed — which is impossible unless the nose is doing something other than matching. And it would predict that a small change to a molecule leaves it recognisable, since the key is nearly the same shape. That prediction fails badly: octanol smells faintly citrus and octanoic acid, differing by an oxidised end group, smells rancid and sweaty. Even the two mirror-image forms of the same molecule part ways — one form of carvone smells of spearmint, its reflection of caraway.

Linda Buck and Richard Axel found the receptor family in 1991, and the second half of the answer came from the same laboratory later: each receptor responds to many different odorants, and each odorant activates many different receptors. No receptor is dedicated. Ask what follows from that and the arithmetic reverses. If a molecule's signature is not which single receptor fired but which subset fired, and how strongly, then the capacity is not 400 — it is the number of distinguishable subsets, which is astronomically larger than the number of parts.

There is a further piece of tidiness that makes the pattern readable. Each sensory neuron in the nose expresses essentially one receptor type, and all the neurons carrying the same type send their wires to the same one or two spots in the olfactory bulb. So the scattered chemistry in the nasal lining is gathered into a fixed map: an odour arrives as a particular constellation of bright spots on a layout that is the same every time.

Does this predict anything we can check? It predicts that similar molecules should smell related rather than unrelated, because overlapping shapes activate overlapping receptor sets — and broadly they do, which is why chemists can sometimes guess a smell family from a structure. It also predicts something less comfortable: that concentration should matter, since raising the amount recruits weakly-responding receptors and changes the pattern. That is exactly what happens. Indole is faecal in a jar and floral at trace amounts in jasmine.

Three refinements hold the picture together.

The first is that "the code" lives at a level above any single receptor. Knowing that receptor 173 fired tells you almost nothing, in the way that knowing a word contains an e tells you almost nothing. This is why olfaction resisted explanation so long: researchers kept looking for the dedicated detector and there is none.

The second is that combinatorial capacity is not the same as discriminating capacity. It is tempting to multiply out the subsets and announce an enormous number of distinguishable smells; a widely-repeated figure of about a trillion comes from exactly that kind of extrapolation and is genuinely disputed, having been challenged on statistical grounds shortly after it was published. The honest statement is the qualitative one: combination lifts the ceiling far above the parts count, and where the true ceiling sits is unsettled.

The third is that the receptor pattern is the input to perception, not perception itself. What you experience as "coffee" is a hundred-odd volatile compounds arriving together and being read as a single learned object, shaped by memory, expectation and language. The pattern is the message; the brain still has to be a reader.

• Why most of the human odorant-receptor gene family has decayed into pseudogenes, and what that says about our sense of smell relative to a mouse's.
• How the brain learns to treat a hundred simultaneous volatiles as the single object "coffee".

Every term the collection defines is gathered in the glossary.