In 1952, two years before he died, Alan Turing published his only paper on biology. He called it "The Chemical Basis of Morphogenesis." The question it addressed is ancient: how does a spherically symmetric embryo become an asymmetric organism with a head, a tail, fingers, stripes? How does uniform become patterned?
His answer was that it could happen through chemistry alone — no blueprint, no template, no external signal required. Two chemicals, diffusing through tissue and reacting with each other, could spontaneously break symmetry. Starting from near-uniform concentrations with only tiny random fluctuations, the system would develop stable spatial patterns of concentration: peaks and troughs, spots and stripes, organized structure where there had been none.
The part that is hardest to accept is that the force driving this is diffusion.
Diffusion is the agent of erasure. It's what destroys gradients, homogenizes systems, dissipates structure. Drop ink in water and diffusion spreads it to uniform gray. In every other context, diffusion works against spatial pattern. Turing showed that under the right conditions, diffusion creates it. A system perfectly stable without diffusion can be made to generate patterns by adding diffusion back in. The stability analysis calls this, correctly, diffusion-driven instability.
The mechanism requires two molecules: an activator and an inhibitor. The activator is autocatalytic — a small local increase in activator concentration generates more activator, a positive feedback loop. The activator also promotes production of the inhibitor. The inhibitor, for its part, suppresses the activator. So far this is just a standard negative feedback system, and without diffusion it's stable: any local perturbation gets damped.
What breaks the symmetry is the diffusion asymmetry. The inhibitor must diffuse significantly faster than the activator — a ratio of at least 5 to 10 in many biological systems, sometimes 100. When you add this disparity back in, the local behavior changes: a random peak of activator starts to self-amplify, as before. It also starts producing inhibitor, as before. But the inhibitor, being faster, diffuses outward before it can accumulate locally enough to suppress the activator. The inhibitor runs away. It spreads laterally into the surrounding tissue, where it suppresses activator formation at a distance. The original peak is left behind with its self-amplification intact and its local inhibitor depleted — and it keeps growing.
The result is an island of high activator concentration, surrounded by a ring where the inhibitor is elevated and activator is suppressed. Beyond that ring, inhibitor concentration has fallen enough that a new island can form. The spacing between peaks — the characteristic pattern wavelength — is set by the ratio of diffusion rates and the kinetic parameters. The same equations, with different values, produce spots or stripes, close or widely spaced, depending on the parameters and the geometry of the domain.
One of the vivid checks on this theory is the cheetah tail. A cheetah's body is covered in discrete spots. Its tail is ringed with stripes. The tail is narrower than the body — narrow enough that the domain geometry forces the Turing pattern from spots into stripes. The transition from spots to stripes along a single animal, at the point where the domain narrows, is exactly what the equations predict. You can get spots on the body and stripes on the tail from the same parameters, just by changing the width of the domain.
Shigeru Kondo and R. Asai showed in 1995 that a marine angelfish — Pomacanthus — provides a different kind of evidence. As the fish grows, its stripe pattern doesn't simply scale up. The stripes actively rearrange: new stripes insert into the existing pattern to maintain the characteristic spacing. A static growth model would just produce stripes thickening with the fish. What Turing's equations predict, and what the fish shows, is dynamic pattern insertion — the spacing is preserved by adding new stripes as the domain grows. Kondo's simulation correctly predicted the future rearrangements observed in real fish.
Experimental confirmation in chemistry came in 1990, 38 years after the theory. Castets, Dulos, Boissonade, and De Kepper at the University of Bordeaux used the CIMA reaction (chlorite-iodide-malonic acid) in a thin agarose gel. The gel was critical: it immobilized the starch used as an indicator, which bound to the activator iodide and reduced its effective diffusion rate, creating the required asymmetry. The result was standing spots and stripes in a chemical system — the first unambiguous Turing pattern outside a living organism.
Molecular biological confirmation took longer. In 2006, Sick et al. identified WNT (a signaling protein) as the activator and DKK (its antagonist) as the inhibitor in hair follicle spacing in mice. Reducing DKK expression produced denser follicles; the quantitative predictions matched. In 2012, two independent papers appeared in rapid succession. Economou et al. showed that the transverse ridges on the mammalian palate form via FGF (activator) and Sonic hedgehog (inhibitor), and identified the diagnostic Turing signature: when a ridge was surgically removed, the gap was filled not by new ridge growth at the wound edge but by bifurcation of neighboring ridges — the pattern always inserts new peaks by splitting existing ones. And Sheth, Marcon, Bastida, Sharpe, Ros, and others published direct evidence in digit patterning: Hox gene dosage controls the wavelength of the Turing pattern — reduce Hox gene expression progressively and the predicted digit-forming stripes become thinner and more numerous, eventually bifurcating. The fingers, as they form in the mouse embryo, are a Turing pattern. The Hox genes are tuning its wavelength.
That's 54 years from theory to molecular confirmation for that particular result. Turing published the paper in August 1952 and was dead by June 1954. He didn't know about the 1990 chemistry experiment. He didn't know about the angelfish, the hair follicles, the palate ridges, the digit stripes. He worked it out from linear stability analysis and the theory of partial differential equations, applied to a question no mathematician had seriously attempted before. He had no experimental evidence. He had only the mathematics suggesting that this was how it could work.
The reason to find this satisfying — beyond the specifics of morphogenesis — is what it shows about the diffusion-driven instability itself. The force that destroys spatial structure is the same force that generates it, under a particular asymmetry. The inhibitor running away is not incidental to the mechanism; it is the mechanism. Without that lateral escape, the inhibitor would suppress the local peak before it could establish, and the system would return to uniformity. The structure forms because the inhibitor doesn't stay.
Diffusion doesn't pattern things despite being a smoothing force. It patterns things through a specific exploitation of its smoothing: the inhibitor diffuses into homogeneity and leaves the activator's local positive feedback uncontested. The pattern is built by what runs away.