Skip to main content

Animal Coat Patterns

Module 4: Space and Pattern Formation

NoteReading
  • Compulsory: Murray (1988).
  • Recommended: Kondo and Miura (2010), the part on Turing patterns in vertebrate skin, and Britton (2003), Appendix C, Example C.10 (eigenvalues on a cylinder).

The classical biological application of Turing’s mechanism is Murray’s model of mammalian coat markings (Murray 1981; 1988; 2003, sec. 3.1). This page lists its main claims and uses your 1D tools to reproduce the central one: domain size controls the pattern.

The Model in Brief

  • A reaction-diffusion system sets up a spatial prepattern of a morphogen on the embryo’s skin during a short window of development (Murray 1981).
  • Pigment cells later read the prepattern: where the morphogen exceeds a threshold, they produce melanin (Murray 1981).
  • Only the scale and geometry of the domain were varied in Murray’s simulations. Very small domains stay uniform, and increasingly complex patterns, from stripes to spots, appear as the domain grows. The size and shape of the embryo when the mechanism operates therefore decide the marking (Murray 1988; 2003, sec. 3.1).

The Tail Prediction

A tail is close to a tapering cylinder. Its surface is periodic around the circumference and has zero flux at the two ends. The Laplacian eigenproblem on such a surface is solved by separation of variables, with eigenfunctions \(\cos(n\pi z/h)\,e^{im\phi}\) and eigenvalues

\[ k_{m,n}^2 = \frac{m^2}{r^2} + \frac{n^2\pi^2}{h^2}, \qquad m, n = 0, 1, 2, \dots, \tag{1}\]

where \(r\) is the radius and \(h\) the length of the cylinder (Britton 2003, Appendix C, Example C.10). Modes with \(m = 0\) do not depend on the angle, so they are rings around the tail. Modes with \(m \ge 1\) vary around the circumference, and the combinations with \(n \ge 1\) give spots.

A mode can grow only if \(k_{m,n}^2\) lies in the unstable band. Because \(k_{m,n}^2 \ge m^2/r^2\), no mode with \(m \ge 1\) is unstable when \(1/r\) exceeds the upper edge \(k_2\) of the band, that is when \(r < 1/k_2\). For Case A the band is \(0.349 < k < 0.523\), so the critical radius is \(1/k_2 \approx 1.91\). A thinner cylinder carries only ring-shaped bands. Figure 1 shows this directly.

Figure 1: Circumferential mode numbers \(m\) that have at least one unstable axial partner \(n\), as a function of the cylinder radius \(r\), for Gierer-Meinhardt Case A on a cylinder of length \(h = 40\) with zero flux at the ends. For \(r\) below about \(1.91\) only \(m = 0\) is unstable (rings around the tail). The dashed line marks \(r = 1/k_2\).

The model therefore predicts that a spotted animal can have a striped tail, but a striped animal cannot have a spotted tail. The tail is striped near the tip, where \(r\) is small, and the full pattern is possible only where the radius is large enough (Murray 1988; 2003, sec. 3.1). The argument is linear: it says which modes can start to grow at a given radius, not which pattern the nonlinear terms finally select.

Domain Size in 1D

The same effect appears in 1D. Figure 2 shows the leading unstable Neumann mode for Case A (\(a = 0.4\), \(b = 1\), \(d = 30\), \(\gamma = 1\)) as the domain length grows.

Figure 2: Leading unstable Neumann mode \(n\) versus domain length \(L\) for Gierer-Meinhardt Case A, using the exact Neumann spectrum \(k_n = n\pi/L\). \(n = 0\) means no unstable mode, so the flat state persists. The dashed line is \(L\,k_{\max}/\pi\) for the continuous fastest-growing wavenumber.

Below \(L \approx 6\) and for \(9 < L < 12\) nothing grows. As \(L\) increases the leading mode jumps by one at a time, and the pattern wavelength stays close to \(2\pi/k_{\max} \approx 14.6\). In linear theory, a larger domain therefore holds more peaks at the same spacing rather than wider peaks.

Evidence and Limitations

  • Kondo and Miura (2010) credit Kondo and Asai (Kondo and Asai 1995) with the first observation of the dynamic properties of a Turing pattern in nature, in the stripes of the angelfish Pomacanthus imperator. They report that this dynamic behaviour is shared by many fish species, including the zebrafish, where the stripes rearrange slowly after laser ablation of pigment cells and keep their spacing, as a simulation predicts. See Evidence for Reaction-Diffusion in Real Systems.
  • Kondo and Miura (2010) stress that a simulation that reproduces a real pattern is not proof that the mechanism operates. One has to identify a dynamic property that the simulation predicts and test it, and to identify the interacting components.
  • Linear theory predicts the onset wavelength only. The final pattern also depends on nonlinear terms, domain growth, curvature, and timing (Murray 2003, sec. 2.6 and 3.1).
NoteExercise

Optional extension. Run the 1D solver with \(L = 10\), Neumann boundaries, \(dx = 1\) and \(dt = 0.01\) for Case A. You will see a pattern, although Figure 2 predicts none. Repeat with \(dx = 0.25\) and \(dt = 0.001\). Explain the difference with the effective domain length of the copy boundary condition on Boundary Conditions.

Copying \(u_0 = u_1\) and \(u_{N-1} = u_{N-2}\) places the zero-flux boundaries halfway between the first two and the last two grid points. The effective length is \((N-2)\,dx\). With \(dx = 1\) that is \(8\), and mode \(1\) has \(k = \pi/8 \approx 0.39\), inside the unstable band. With \(dx = 0.25\) it is \(9.5\), \(\pi/9.5 \approx 0.33\) lies below the band, and the flat state persists. This is a discretization effect that disappears under grid refinement, which is why you run the convergence check.

Exit question

NoteExercise

Case A on a Neumann interval of length \(L = 10\) satisfies all four Turing conditions. Will a pattern form? Explain in one or two sentences.

No. The Turing conditions do not depend on the domain, but the admissible wavenumbers \(n\pi/10\) for \(n = 1, 2\) are \(0.31\) and \(0.63\), both outside the unstable band \(0.349 < k < 0.523\). No mode grows, so the flat state persists.

References

Britton, Nicholas F. 2003. Essential Mathematical Biology. Springer Undergraduate Mathematics Series. Springer. https://doi.org/10.1007/978-1-4471-0049-2.
Kondo, Shigeru, and Rihito Asai. 1995. “A Reaction-Diffusion Wave on the Skin of the Marine Angelfish Pomacanthus.” Nature 376: 765–68. https://doi.org/10.1038/376765a0.
Kondo, Shigeru, and Takashi Miura. 2010. “Reaction-Diffusion Model as a Framework for Understanding Biological Pattern Formation.” Science 329 (5999): 1616–20. https://doi.org/10.1126/science.1179047.
Murray, James D. 1981. “A Pre-Pattern Formation Mechanism for Animal Coat Markings.” Journal of Theoretical Biology 88 (1): 161–99. https://doi.org/10.1016/0022-5193(81)90334-9.
Murray, James D. 1988. “How the Leopard Gets Its Spots.” Scientific American 258 (3): 80–87. https://doi.org/10.1038/scientificamerican0388-80.
Murray, James D. 2003. Mathematical Biology II: Spatial Models and Biomedical Applications. 3rd ed. Vol. 18. Interdisciplinary Applied Mathematics. Springer. https://doi.org/10.1007/b98869.