Morphogenesis and Turing Patterns

1. Turing's 1952 paper and the morphogen idea

Alan Turing's paper "The Chemical Basis of Morphogenesis" (1952) addressed one of developmental biology's deepest puzzles: how does a fertilised egg — a single, roughly spherically symmetric cell — give rise to a complex organism with precise spatial organisation? Turing proposed that the answer lay in morphogens: diffusible chemical substances whose concentration gradients provide positional information to cells.

Turing's crucial insight was that a system of two morphogens interacting chemically while diffusing through tissue could be unstable to small spatial perturbations, even if each individual process (diffusion alone, or uniform reaction alone) would be stable. This diffusion-driven instability, now called the Turing instability, is counterintuitive: most people expect diffusion to be a smoothing process that destroys patterns. Turing showed it can instead create them.

Turing's framework was purely abstract — he did not identify specific morphogen molecules. The first confirmed example of a Turing mechanism operating in vivo was the patterning of mouse hair follicles, demonstrated by Sick et al. (2006) who identified WNT and DKK proteins as the activating and inhibiting morphogens. Since then, Turing mechanisms have been confirmed in digit patterning (Raspberry, 2012), palate ridges, and zebrafish stripe formation.

2. Reaction-diffusion equations

The general framework for Turing patterns consists of two coupled partial differential equations describing the concentrations u (activator) and v (inhibitor) in space and time:

The spatial term ∇²u represents Fickian diffusion spreading the concentration in proportion to local curvature. In 1D, ∇²u = ∂²u/∂x²; in 2D it becomes ∂²u/∂x² + ∂²u/∂y². The reaction terms f and g can take many specific forms depending on the model chosen.

A key requirement for a Turing instability is differential diffusion: the inhibitor must diffuse significantly faster than the activator. The biological intuition is that local self-activation (the activator promotes its own production) is balanced by long-range inhibition (the inhibitor diffuses away rapidly and suppresses activator production at a distance), creating a standing wave of chemical concentration.

3. The Turing instability

To understand why diffusion can destabilise a uniform steady state, consider a system at equilibrium (u₀, v₀) that satisfies f(u₀, v₀) = 0 and g(u₀, v₀) = 0. The stability of this state to small perturbations is determined by linearising the system. Writing u = u₀ + ũ and v = v₀ + ṽ, the linearised system becomes:

Substituting a spatial Fourier mode ũ, ṽ ∝ e^(σt + ikx), the system reduces to an eigenvalue problem. The growth rate σ of wavenumber k is:

The Turing conditions require:

1. The uniform state is stable without diffusion: tr(A) < 0 and det(A) > 0.
2. Diffusion destabilises some wavenumber k*: h(k²) < 0 for some k.
3. The second condition requires f_u > 0 (activator self-activates) and g_v < 0 (inhibitor self-inhibits), with d = D_v/D_u sufficiently large.

The critical wavenumber k* at which growth is fastest determines the wavelength of the pattern: λ = 2π/k*. This is the spatial period of stripes or the diameter of spots.

4. Activator-inhibitor systems

The most biologically interpretable version of a Turing system is the activator-inhibitor framework, first explicitly formulated by Gierer and Meinhardt (1972). In this framework:

• Activator (u): a slowly diffusing molecule that promotes its own production and also promotes inhibitor production.
• Inhibitor (v): a rapidly diffusing molecule that suppresses activator production.

The resulting dynamic has a clear spatial logic: where the activator is locally elevated, it amplifies itself (positive feedback) while simultaneously sending the inhibitor diffusing outward. The inhibitor returns the signal, suppressing activation in the region surrounding the peak. This local activation / lateral inhibition (LALI) motif generates a characteristic spacing between pattern elements.

Mathematically, the Jacobian of an activator-inhibitor system at the steady state has the sign structure:

Note that this sign structure means the individual kinetics (without diffusion) make the steady state stable — a state that Turing called conditionally stable. It is the addition of differential diffusion that tips this stable state into instability.

5. The Gierer-Meinhardt model

The Gierer-Meinhardt (GM) model (1972) provides specific reaction kinetics for an activator-inhibitor system, originally inspired by the regeneration of the hydra's head and foot. The standard form is:

The key nonlinearity is the ratio u²/v: activator production is proportional to the square of the activator concentration (autocatalysis) divided by the inhibitor concentration. This nonlinear term creates the saturation needed to stabilise the pattern at finite amplitude.

The GM model successfully reproduced:

• Hydra head and foot patterning and regeneration of excised segments.
• Periodic pigment patterns along fish body stripes and spots.
• Tentacle spacing in hydroid polyps.
• The formation of a single organising centre (head) after random initial perturbations.

6. The Gray-Scott model

The Gray-Scott model (1984), originally developed for chemical reactors, became one of the most studied reaction-diffusion systems due to its extraordinary richness. It models an autocatalytic chemical reaction where substance U is converted to V in the presence of V (autocatalysis), while V slowly decays:

The reaction term u·v² represents an autocatalytic step: each molecule of V "catalyses" converting one molecule of U to V, so the rate grows quadratically with V concentration. The feed term F·(1−u) replenishes U to a uniform baseline.

The Gray-Scott model exhibits an astonishing variety of patterns depending on the (F, k) parameter space:

The "mitosis" mode is particularly striking: localised spots spontaneously elongate and divide into two spots, mimicking cell division. This self-replication property was connected to the Grey-Scott system as a model for self-replicating chemical structures.

7. Biological patterns and experiments

Animal coat patterns

Among the most celebrated applications of Turing's theory is the explanation of animal coat patterns. Murray (1988) demonstrated that by solving reaction-diffusion equations on domains of different sizes and shapes, one can reproduce the patterns observed on animals:

• Leopard spots: typical Turing patterns on a large 2D domain; the spots arise because the domain comfortably fits multiple pattern wavelengths in both dimensions.
• Cheetah/jaguar spots: more irregular, pointing to higher-order or additive contributions of multiple modes.
• Zebra stripes: stripes form on elongated domains because the domain geometry constrains patterns to be oriented transversely.
• Spotted tail of a striped animal: if the tail is thin and cylindrical, the Turing patterns on the tail become rings and spots rather than longitudinal stripes — explaining why many striped animals (e.g. tigers) can have spotted or ringed tails.

Zebrafish stripes: confirmed Turing mechanism

The clearest confirmed Turing mechanism in vertebrate pigmentation is zebrafish stripe formation, where three types of pigment cell (melanophores, xanthophores, and iridophores) interact in a pattern consistent with a multi-component reaction-diffusion system. Nakamasu et al. (2009) showed experimentally that melanophores activate neighbouring xanthophores (short-range activation) while xanthophores inhibit distant melanophores (long-range inhibition), satisfying the LALI requirement.

Mouse hair follicle spacing

Sick et al. (2006, Science) provided perhaps the best direct molecular evidence for a Turing mechanism. They showed that WNT (Wingless/Int signalling) acts as the activator and DKK1 (Dickkopf-1, a WNT inhibitor) acts as the inhibitor in determining the spacing of hair follicles in mouse skin. Perturbing DKK1 expression altered the spacing exactly as Turing theory predicted.

8. JavaScript 2D reaction-diffusion simulation

The interactive simulation below evolves the Gray-Scott model on a 128 × 128 grid using finite differences. Each pixel represents a grid point; colour encodes the concentration of species V (the product). Change the feed and kill parameters to explore the rich pattern phase diagram — from spots to stripes to labyrinthine mazes. Click the canvas to add a drop of V and initiate pattern growth from scratch.

The simulation uses a ping-pong buffer technique: two arrays store the current and next state, alternating each step to avoid read-write conflicts. The Laplacian is computed using the standard 5-point finite difference stencil with periodic boundary conditions. Each animation frame advances multiple reaction-diffusion steps to compensate for the small dt required for numerical stability.

// Gray-Scott parameters
const Du = 0.20, Dv = 0.10;
let F = 0.037, k = 0.060;
const dt = 1.0;

function step(u, v, u2, v2, N) {
  for (let y = 0; y < N; y++) {
    for (let x = 0; x < N; x++) {
      const i = y * N + x;
      // Periodic neighbour indices
      const xp = (x + 1) % N, xm = (x - 1 + N) % N;
      const yp = (y + 1) % N, ym = (y - 1 + N) % N;

      // 5-point Laplacian
      const Lu = u[ym*N+x] + u[yp*N+x] + u[y*N+xm] + u[y*N+xp] - 4*u[i];
      const Lv = v[ym*N+x] + v[yp*N+x] + v[y*N+xm] + v[y*N+xp] - 4*v[i];

      const uvv = u[i] * v[i] * v[i];  // autocatalytic term
      u2[i] = u[i] + dt * (Du * Lu - uvv + F * (1 - u[i]));
      v2[i] = v[i] + dt * (Dv * Lv + uvv - (F + k) * v[i]);
    }
  }
}