loading...
Experiment result (A: free-running period against turnover d; B: peak interval against light cycle length)
running...
How to use
Pick a preset or move the sliders: the model runs live. Orbit with the mouse (drag), zoom with the wheel; the camera buttons fly to the cell, the nucleus, the Per gene, the SCN, the dial and the gauge tubes, and Tour cycles through them. Red particles are Per mRNA leaving the nucleus through the pores, amber particles are PER protein made at the ribosomes, blue particles are the PER:CRY repressor inside the nucleus; those sitting on the green promoter are the bound fraction that switches the gene off. The lamp is on during the light half of the cycle; Light pulse gives a one hour flash. Space pauses, R resets. The 2D page is one click away (Switch to 2D). The number of particles is proportional to the concentrations (a display scale, not a molecule count).
The science
A circadian clock is a gene-protein loop that takes about a day to go round. In mammals the activators CLOCK:BMAL1 switch on the Per and Cry genes; the PER and CRY proteins build up, enter the nucleus and shut CLOCK:BMAL1 down, then are degraded, and the cycle restarts. Each cell of the suprachiasmatic nucleus (SCN, about 20,000 neurons just above the optic chiasm) runs such a loop, and light reaching the retina adjusts it.
The model. This is the classic Goodwin (1965) three-stage negative-feedback loop, with the same equations, parameters and Euler step (at most 0.02 h) as the 2D page:
dX/dt = 10 (1 + ls L(t)) / (1 + Zn) - d X (Per mRNA)
dY/dt = d (X - Y) (PER protein, cytosol)
dZ/dt = d (Y - Z) (nuclear PER:CRY repressor)
L(t) is 1 in the light (a 1 h pulse, or the first half of each light cycle) and 0 otherwise, ls is the light strength; all quantities are in arbitrary units, time is in hours. Initial state X = 0.1, Y = 0.2, Z = 2.5.
Why it oscillates. The fixed point X = Y = Z = Z* solves 10 / (1 + Z*n) = d Z*. Linearising, three first-order stages with equal rates d and one inhibitory feedback of gain g = n Z*n / (1 + Z*n) are unstable only if g > 8 (secant condition), which is why a Hill coefficient n of at least 9 is needed here (Griffith 1968). Near onset the angular frequency is √3 d, i.e. a period 2π/(√3 d) = 22.96 h for d = 0.158 /h; the full nonlinear loop gives 24.9 h because the oscillation is far from sinusoidal.
Entrainment. A periodic light drive pulls the clock toward its own period when the mismatch is small enough (an Arnold tongue); Experiment B measures the interval between mRNA peaks for light cycles from 18 h to 30 h and shows where the clock locks (the points sit on the diagonal) and where it does not.
Numerical checks run in the page (separate code paths from the live engine): (1) in a damped regime (n = 4) the integrated state reaches the closed-form Z*; (2) the engine period (Euler, 0.02 h) against an independent 4th-order Runge-Kutta run at 0.005 h with interpolated peaks (about 1% apart, tolerance 3%); (3) the scaling T(1.25 d) / T(d) = 0.8; (4) the linear Hopf estimate, which is expected to be a rough guide only; (5) the secant condition predicts correctly whether the loop oscillates; (6) the reference entrainment case locks to exactly 24 h.
Honest simplifications. (a) A Hill coefficient of 10 is not a real binding cooperativity; in real clocks the steepness comes from extra delays, phosphorylation, multimer formation and several interlocked loops (REV-ERB/ROR, Cry/Per). (b) Time is in hours but the units of X, Y, Z and the constant 10 are arbitrary; the 24 h period is a tuning of d, not a prediction. (c) One oscillator, no noise, no SCN coupling (VIP signalling), no temperature compensation. (d) The 3D scene is schematic: particle counts follow the concentrations, they are not a stochastic simulation; the SCN glow shows the mRNA level, and the retina-to-SCN pathway is drawn only to show where light enters. (e) The light response (a multiplier on transcription) is the simplest possible and phase shifts are not calibrated to human data.
Sources to spot-check (quoted from memory, flagged): Goodwin 1965 (Adv Enzyme Regul 3); Griffith 1968 (J Theor Biol 20); Leloup and Goldbeter 2003 (PNAS 100); Reppert and Weaver 2002 (Nature 418); Takahashi 2017 (Nat Rev Genet 18). Educational model, not a medical tool.