Natural Patterns: Spots, Stripes, and Seashells

Animal Coat Patterns: The Classic Examples

James Murray’s analysis of mammalian coat patterns, developed through the 1980s and presented comprehensively in Mathematical Biology (1989, 2003), applied the Turing mechanism to a simple question: why do some animals have spots and others have stripes?

Murray’s argument begins with a geometric observation. Reaction-diffusion equations on a domain produce patterns whose spatial structure depends on the domain’s size and shape relative to the pattern wavelength. On a domain that is large in both dimensions (a torso), the full two-dimensional instability operates: both spot patterns (hexagonal arrays) and stripe patterns are possible, with the specific type depending on kinetic parameters. On a domain that is narrow in one dimension (a tail or a leg), only the long-axis spatial mode fits — the transverse dimension is too small to support a full wavelength. The result is stripes aligned perpendicular to the long axis.

This yields a testable prediction: animals with spotted bodies may have striped tails, but no animal has a spotted tail with a striped body. The logic is geometric — the tail is narrower than the body, and a narrow domain can only support stripes. Murray surveyed mammals and found the prediction confirmed: cheetahs have spotted bodies and striped tails; leopards have spotted bodies and ringed tails (rings being stripes wrapped around a cylinder); no counterexample has been documented.

A second prediction: very small animals (mice, small rodents) should not have coat patterns at all, because the body is too small relative to the pattern wavelength to support any instability. Very large animals should not have patterns either, because the pattern wavelength becomes small relative to the body, producing patterns too fine to be visible. Murray argued that coat patterns are most prominent in medium-sized mammals, where the body is the right size relative to the wavelength — consistent with observation.

These predictions are necessary but not sufficient to confirm the Turing mechanism. A domain-geometry argument that correctly predicts where stripes and spots appear does not, by itself, prove that the patterning mechanism is reaction-diffusion rather than some other process that is also geometry-dependent. The argument is consistent with the Turing mechanism, but confirmation requires molecular evidence.

The Molecular Evidence: Skin and Hair Follicles

The critical shift from mathematical modeling to molecular biology came through the study of zebrafish (Danio rerio) pigmentation.

Zebrafish have horizontal dark stripes formed by melanophore cells (dark pigment) alternating with light inter-stripe regions populated by xanthophore cells (yellow pigment) and iridophore cells (iridescent). Shigeru Kondo’s group at Osaka University investigated the interaction logic between these cell types over two decades of experimental work.

The key findings: melanophores and xanthophores repel each other at short range — when placed adjacent, they move apart. At longer range, each cell type promotes the survival and differentiation of the other — melanophore survival requires signaling from distant xanthophores, and vice versa. This interaction structure is precisely the activator-inhibitor architecture: short-range activation (cells of the same type reinforce each other’s presence locally) coupled with long-range inhibition (cells of the opposite type suppress each other at a distance, mediated by the faster-spreading signals).

Kondo’s 2009 paper in PNAS, “Interactions Between Zebrafish Pigment Cells Responsible for the Generation of Turing Patterns,” demonstrated that the cell interaction network has the mathematical properties required for Turing instability. Crucially, the “diffusion” in this system is not molecular diffusion of a chemical — it is the effective spreading of influence through cell projections and signaling molecules. The cell-based implementation of the Turing mechanism operates through different physical substrates than Turing imagined, but the mathematical structure is identical.

Experimental perturbations confirmed the model. Ablating melanophores from a stripe region caused the stripe to regenerate — consistent with the Turing mechanism’s prediction that the pattern is dynamically maintained, not laid down once and frozen. Overexpressing certain signaling molecules shifted the stripe width in the predicted direction. Mutant zebrafish with altered melanophore-xanthophore interactions (e.g., the leopard mutant) produce spots instead of stripes — a pattern change consistent with shifting parameters in the Turing model.

In mouse hair follicle spacing, Sick et al. (2006) identified WNT signaling as the activator and DKK (Dickkopf) as the inhibitor. WNT promotes its own expression (autocatalysis) and promotes DKK expression. DKK inhibits WNT signaling and diffuses faster than the WNT signal. Overexpressing WNT increased follicle density (shorter pattern wavelength); overexpressing DKK decreased it — both consistent with Turing model predictions. The paper, published in Science, was one of the first to identify specific molecular candidates for the activator and inhibitor in a mammalian Turing system.

Digit Patterning and Limb Development

The most dramatic confirmation of the Turing mechanism in vertebrate development came from digit patterning. The standard model of limb development for decades was Wolpert’s positional information framework: digits form at positions specified by a gradient of Sonic hedgehog (Shh) morphogen emanating from the zone of polarizing activity at the posterior edge of the limb bud. Cells read their position along the gradient and differentiate accordingly.

Sheth et al. (2012) challenged this model. Working with mouse limb buds, they showed that reducing the dosage of Hox genes — which modulate the parameters of pattern formation — changed the number and spacing of digits in a way consistent with Turing predictions but inconsistent with a simple gradient model. Reducing Hox gene dosage produced extra digits with reduced spacing, as if the pattern wavelength had shortened. This is precisely what the Turing model predicts when the “activator” range decreases relative to the “inhibitor” range.

Raspopovic et al. (2014) went further, identifying a specific Turing network in the distal limb bud: BMP (bone morphogenetic protein) acts as the activator, promoting its own expression and promoting Sox9 (a chondrogenic transcription factor). WNT acts as a modulator. The resulting network has the topology of a Turing system, and perturbations to BMP diffusion or Sox9 sensitivity shifted digit number and spacing in quantitative agreement with the model.

The implication is substantial. Digit patterning — one of the most studied problems in developmental biology — appears to involve a Turing mechanism operating within a morphogen-gradient framework. The two mechanisms are not alternatives; they are layered. The gradient (Shh) sets the overall domain and modulates parameters. The Turing instability, operating within that domain, generates the periodic digit primordia. This layered architecture resolves a long-standing puzzle: the gradient provides robustness and global polarity (which Turing patterns lack), while the Turing mechanism provides the periodicity and the capacity to adjust digit number in response to limb-bud size.

Seashell Pigmentation: Meinhardt’s Models

Hans Meinhardt spent decades modeling the pigmentation patterns on mollusc shells, producing a body of work collected in The Algorithmic Beauty of Sea Shells (first edition 1995, fourth edition 2009). His approach exploited a simplifying feature of shell growth: the mantle (the tissue that secretes the shell) is a thin line at the shell’s growing edge, and the pigment pattern is laid down as a one-dimensional trace at the mantle edge over time. The shell surface is thus a spacetime diagram: one spatial dimension (around the mantle edge) and one temporal dimension (as the shell grows).

This reduction from two-dimensional pattern formation to a one-dimensional reaction-diffusion system running in time makes shell patterns particularly tractable. Stripes that run parallel to the growth direction are steady-state spatial patterns — the reaction-diffusion system at the mantle edge is in a fixed patterned state that does not change as growth proceeds. Stripes perpendicular to the growth direction indicate temporal oscillation — the mantle alternates between pigmented and unpigmented states. Oblique lines, zigzags, and branching patterns indicate traveling waves, wave collisions, and more complex spatiotemporal dynamics.

Meinhardt showed that a standard activator-inhibitor model operating on a one-dimensional domain (the mantle edge), with its output recorded as the shell grows, reproduces a remarkable catalog of real shell patterns. Different parameter regimes produce different shell-pattern types, and the correspondence between simulated and real patterns is often striking. Meinhardt’s model predicts that certain pattern combinations are forbidden — for example, a shell cannot have oblique lines that change direction without a branching or collision event — and these constraints are observed in real shell taxonomy.

The shell pattern work is particularly valuable because the spacetime interpretation makes the dynamics directly visible. A photograph of a cone shell’s surface is a record of the reaction-diffusion dynamics at the mantle edge over the shell’s entire growth history. The pattern is not just a static structure — it is a complete dynamical trace, readable to anyone who knows the model.

The limitation is that molecular identification of the activator and inhibitor in mollusc pigmentation has been less successful than in vertebrates. The patterns are consistent with a reaction-diffusion mechanism, and no alternative mechanism has been proposed that reproduces the full range of observed patterns, but the specific molecules involved remain largely unidentified. The evidence is structural (pattern-level agreement) rather than molecular.

Further Reading