Lateral inhibition in tissue

Why does a patch of cells that are all equally able to become a bristle end up producing bristles at neat, even spacings?

Look at the back of a fly. The bristles are neither random nor crowded — they stand apart at fairly regular intervals, each surrounded by ordinary skin. Look at the tissue before they appear and you find a patch of cells all expressing the same genes, all equally capable of becoming a bristle. None is marked.

The tempting explanation is that something told each cell which job to take. But if a coordinate system were doing the work, what would read it, and how would a sheet a few dozen cells across encode position finely enough to pick out every third or fourth cell? Could identical cells generate the spacing themselves?

Suppose you wanted a rule producing spacing with no map. What is the least it would have to say? Something like: whichever of us takes the job, the ones beside it must not. Local, negative, no coordinates and no census — just a prohibition passed to neighbours.

So let each cell send an inhibiting signal to the cells it touches. Immediately there is a problem: if all cells are identical they send and receive equally, and nothing happens. What breaks the tie? Nothing more than noise — gene expression is stochastic, so at any instant one cell happens to be making slightly more signal than its neighbour. That difference is tiny and would ordinarily wash out, so the mechanism needs a second property: it must amplify the difference rather than damp it.

Here is the arrangement that does it, and it is worth building carefully because the logic is easy to get backwards. Every cell carries both the signal and the receptor for it. In animals these are Delta, a protein on the cell surface, and Notch, the receptor on the surface of the cell it touches — both membrane-bound, so the signal travels no further than contact allows. When Delta on cell A activates Notch on cell B, the receptor's inner portion is cleaved off, enters B's nucleus, and switches on repressors — the HES family, or E(spl) in the fly — which shut down the genes that make B neural, including B's own Delta.

Trace that. More Delta in A means less Delta in B. Less Delta in B means less inhibition arriving at A, letting A make more Delta. A signal that is inhibitory between cells becomes, once round the loop twice, self-reinforcing within one. Two negatives make a positive, and the chance difference grows instead of fading, until one cell is saturated with Delta and its neighbours with active Notch — one bristle precursor, a ring of plain skin, and no map anywhere.

What sets the spacing? The range of the signal, and nothing else. Because Delta and Notch are stuck in membranes, the prohibition reaches roughly one cell diameter. Real spacings are often wider, and the reason is that in some tissues — fly bristle patterning is the studied case — cells extend long thin filopodia carrying Delta several diameters away, lengthening the reach of the inhibition and widening the pattern to match. Change the reach, change the pitch.

Now the part that resists the adaptive telling. Which cell wins? Essentially whichever was ahead when amplification began — not the best-placed, or the healthiest, or the one nearest anything. The pattern is regular; the identity of its members is not settled in advance. In the fly this shows plainly: the large bristles sit in positions reproducible between animals, because a separate prepattern confines the competing clusters to particular small regions, but the small bristles elsewhere are only statistically regular, differing between individuals and even between the two sides of one fly.

Two comparisons sharpen what this mechanism is and is not.

It is not a gradient. Positional information — a diffusing morphogen read as a coordinate — also produces reliable patterns, but it needs a source, a decay, and thresholds calibrated against distance. Lateral inhibition needs none of those. It produces spacing without ever specifying position, which is why it can pattern a tissue with no landmark to measure from.

Nor is it quorum sensing, though both are cell-to-cell signalling with feedback. Bacteria use a diffusible signal reaching everyone alike, to make a whole population do the same thing at once. This uses a contact-bound signal, to make touching cells do different things. Diffusible converges, contact-bound diverges, and the difference in the resulting pattern follows directly from how far the molecule can travel.

The account is sharply falsifiable, and was in fact falsified-in-principle first and explained second. If mutual inhibition confines the fate to one cell, removing the signalling should let every cell in the cluster take it. That is exactly what a class of fly mutants does — lose Notch or Delta and far too many cells become neural, which is why these were named the neurogenic genes long before the logic above was worked out. Force Notch permanently active and you get none. A clone unable to make Delta should be inhibited while inhibiting nobody, and should lose. The refuting observation is clean: a cluster with Notch signalling abolished that still produced exactly one bristle would leave the model nothing to stand on, since the singling-out would have to come from somewhere else entirely.

• How the fly's large bristles get reproducible positions, and what a prepattern contributes on top of this mechanism.
• Whether the same Notch circuit used between cells that are not equivalent to begin with is the same mechanism or a different one wearing the same proteins.

Every term the collection defines is gathered in the glossary.