Biology × Mathematics × 1952

How Leopards Got Their Spots

The accidental discovery that a dead mathematician's last idea explains every stripe, spot, and swirl on every animal on Earth

scroll to explore

Here is a thing that seems obvious until you think about it: a leopard's spots are not painted on. They weren't arranged by a designer, they weren't handed down genetically as a blueprint, and nothing measured out the distances between them. The pattern grew. Out of cells. Out of chemistry. Out of nothing but a few molecules interacting.

If you look at a leopard skin long enough, a question starts nagging: why are the spots roughly the same size? Why evenly spaced? Why not a gradient, or a checkerboard, or random noise? Why does the pattern have structure at all?

The answer was worked out in 1952 by Alan Turing — the mathematician who cracked the Enigma code and invented the theoretical basis for all modern computers. It was, improbably, his last great idea. And it turns out to explain not just leopards, but zebras, tropical fish, seashells, coral, human fingerprints, and the spacing of hairs on your skin.

Act I — A Very Strange Idea

Turing's insight was this: you can get a pattern — a regular, spatially organised pattern — from a completely featureless, uniform medium, using nothing but two chemicals that react with each other and spread (diffuse) through space.

One chemical is the activator. It promotes its own production. If you have a little bit of it somewhere, more of it will be made at that location. It's self-amplifying.

The other chemical is the inhibitor. The activator also triggers the production of the inhibitor, and the inhibitor suppresses the activator. So the activator creates its own enemy.

So far, this just sounds like a system that would oscillate or reach some equilibrium. Here is the clever bit — the part that makes it spatial rather than merely temporal:

The inhibitor diffuses much faster than the activator.

Imagine dropping a tiny seed of activator in an empty field. Locally, it amplifies itself — the spot of activator grows. But the inhibitor it generates spreads far outward, suppressing any new activator from forming nearby. The result: a spot of activator surrounded by an inhibitor moat. Other spots can only form where they're far enough away to escape the moat.

activator spreads slowly inhibitor spreads fast → new spots can only form outside the inhibition zone Local activation · Long-range inhibition

This one principle — local activation, long-range inhibition — is all you need. Given enough space and time, a uniform field seeded with a tiny amount of random noise will self-organise into a regular pattern. Spots, stripes, or spirals, depending on the exact rates.

The mechanism that determines which pattern forms is the ratio of how fast each chemical diffuses. Change that ratio, and you change from a leopard to a zebra.

Act II — Watch It Happen

Below is a live simulation. The grid starts with near-uniform chemistry — just a tiny sprinkle of random noise in the activator concentration. You can watch the pattern emerge, completely spontaneously, over the next few seconds.

When it finishes, click anywhere on the canvas to seed new activator at that point. You're playing the role of a mutation, a scratch, a local perturbation — and the chemistry will respond.

Live Reaction-Diffusion — Gray-Scott Model
Initialising... watch for pattern formation
Try it Wait for the pattern to fully develop (~10 seconds), then click or drag on the canvas. Each touch seeds a pulse of activator. In some parameter regimes, new spots will bud off. In others, the perturbation will stabilise, or ripple outward and die. The pattern is not fragile — it actively defends its structure.

What you're watching isn't programmed. No cell knows where it is. No gene says "put a spot here." Every pixel is following the same two rules: activate yourself, generate an inhibitor that travels farther than you do. The global pattern is an emergent property.

Act III — The Dial Between Leopards and Zebras

The equations Turing used have only two free parameters: let's call them the feed rate F (how quickly fresh activator is introduced into the system) and the kill rate k (how quickly the inhibitor decays). Change F and k, and you change everything about the pattern.

This is the remarkable bit. There isn't one set of equations for leopards and a different set for zebras. There's one system. The same two chemicals, the same two rules. Different species simply evolved to different positions in the (F, k) parameter space.

∂u/∂t = Dᵤ·∇²u − uv² + F(1−u)
∂v/∂t = Dᵥ·∇²v + uv² − (F+k)v

u is the activator concentration, v is the inhibitor, Dᵤ and Dᵥ are their diffusion rates (Dᵥ is always larger — that's the key), and ∇² is the Laplacian, which just means "how much does this quantity differ from its local neighbourhood." The term uv² is the heart of it all: the activator is consumed when it meets the inhibitor, in proportion to how much of each is present.

Pick an animal below and watch the pattern regenerate. All of them are the same simulation — only F and k change.

Parameter Explorer — One Equation, Many Animals
Developing...
Presets
FEED RATE F 0.037
KILL RATE k 0.060
Try it Start with 🐆 Leopard, then switch to 🦓 Zebra without waiting — you'll see the existing spots slowly stretch and merge into stripes. The pattern doesn't reset; it transforms. This is what actually happens during embryonic development: as the organism grows, the effective diffusion rates shift, and a pattern that began as one type flows into another.

Notice what happens at extreme values. Very high F (lots of fresh activator) floods the system — you get waves or turbulent activity. Very low k (inhibitor barely decays) freezes the system into a nearly uniform state. The interesting patterns live in a sweet spot between these extremes, which is exactly where biology tends to operate.

There's a beautifully testable prediction hidden here. If stripes and spots are on the same continuum, then animals with intermediate body proportions should show mixed patterns. And they do: cheetahs have spots on their body but stripes on their tail. Genets have spots on their torso and rings on their tail. In the tail — a thin, long tube — diffusion geometry is different, and the same chemistry produces different patterns. The topology of the surface changes the output of the reaction.

Act IV — What Turing Couldn't See

Alan Turing published "The Chemical Basis of Morphogenesis" in March 1952. He was 39. He had no computer capable of running these simulations — he worked out the mathematics by hand, predicting the existence of pattern-forming instabilities purely from the equations. He called the mechanism a diffusion-driven instability, which later came to be known as a Turing instability.

He died in June 1954. The first proper computer simulations of his equations came in the 1970s. The first experimental confirmation in living biology — directly measuring the two chemicals and showing they matched his predictions — didn't arrive until 2012, when researchers at King's College London tracked the molecular signals that space out hair follicles in mouse skin. The chemicals were named Wnt (activator) and Dkk (inhibitor), and they diffused at exactly the ratio Turing's equations required.

Sixty years after a hand-written paper, in an organism he never studied, for a phenomenon he described in the language of pure mathematics: confirmed.

Since then, Turing patterns have been found in the development of teeth spacing, finger and toe spacing, feather distribution in birds, the geometry of human fingerprints, the labyrinthine folds of the gut lining, and the patterning of colour in tropical fish — where they were observed changing in real time as the fish aged, exactly as Turing's model predicts.

A final experiment Go back to Demo 2. Set F very low (~0.015) and k around 0.055. You'll see something remarkable: a pattern of dots that periodically divides — one dot becomes two, each migrates to a new position, and stabilises. This self-replication of spots bears a striking resemblance to cell division, emerging from the same two-chemical system, no additional rules required. Turing didn't set out to model cell division. He modelled chemical diffusion. The biology emerged from the math.

The leopard doesn't know it carries a Turing pattern. The equations don't know what an animal is. Somewhere in between — in the chemical gradients of a developing embryo, in the feedback loops of reaction and diffusion — a blank canvas acquires structure. The same way a flat pool of water, given the right wind, produces regular ripples. Not because someone drew them. Because the physics had no other choice.


Alan Turing, "The Chemical Basis of Morphogenesis," Phil. Trans. Royal Soc. B, 1952.
Sheth et al., "Hox genes regulate digit patterning by controlling the wavelength of a Turing-type mechanism," Science, 2012.
Yamaguchi et al., "Pattern regulation in the stripe of zebrafish suggests an underlying dynamic and autonomous mechanism," PNAS, 2007.
The Gray-Scott model used here was characterised by John Pearson in Science, 1993.