Brownian Motion Inside the Cell — Fluctuation-Dissipation at the Cellular Scale

1. Why the Cell Is a Different Physical Regime

The fluctuation-dissipation theorem establishes, in general, that any system in thermal equilibrium exhibits random fluctuations whose statistics are locked to the same friction coefficient that damps its response to a force — the theorem itself, its derivation via the Langevin equation, and the Green-Kubo formalism are covered there in full and are not repeated here. What changes inside a cell is not the theorem but the numbers you plug into it: the effective viscosity, the size of the diffusing particle, and — critically — the assumption that the medium behaves like a simple Newtonian fluid at all.

At the scale of a bacterium (1-2 µm) or a mammalian cell (10-30 µm), inertia is irrelevant. The Reynolds number for a protein moving at diffusive speeds is of order 10⁻⁶ — viscous forces dominate inertial forces by six orders of magnitude. A swimming bacterium that stops paddling halts within a fraction of a nanosecond; there is no coasting. This is Purcell's "life at low Reynolds number," and it means every displacement inside the cell is set by a balance between thermal kicks and drag, exactly the fluctuation-dissipation balance, just evaluated in an intensely crowded medium.

2. The Stokes-Einstein Relation at Micron Scale

For a spherical particle of radius r in a fluid of viscosity η, the diffusion coefficient is:

Cytoplasm's bulk viscosity is roughly 1.5-4× that of water for small solutes, but the effective viscosity felt by a particle rises steeply with its size, because larger objects experience the full mesh of the cytoskeleton and macromolecular crowd, not just the solvent. Rough orders of magnitude for a eukaryotic cell at 37°C:

A useful rule of thumb from the Einstein relation in 3D, ⟨r²⟩ = 6Dt, is that a typical soluble protein crosses a 10 µm cell by free diffusion in well under a second, which is why the cell rarely needs active transport for small-molecule signalling — but why it absolutely does for anything larger, as §5 explains.

3. Macromolecular Crowding

Cytoplasm is not a dilute protein solution — it is 20-30% volume fraction occupied by proteins, ribosomes, RNA and cytoskeletal filaments, comparable to a saturated protein crystal. This macromolecular crowding has two distinct effects that a naive Stokes-Einstein calculation misses entirely:

• Excluded-volume thermodynamics: crowding agents don't bind a tracked molecule, but they exclude it from the volume they occupy, which raises the effective concentration and reaction rate of everything else — sometimes called the "colloidal osmotic pressure" of the cytoplasm.
• Obstruction of motion: a diffusing particle must navigate around a dense, transient mesh of obstacles, which slows long-range diffusion far more than short-range diffusion.

The consequence is a diffusion coefficient that depends on the length- and time-scale of observation — D(t) rather than a single constant D — and it is the direct cause of the anomalous scaling discussed next.

4. Anomalous Sub-Diffusion of Vesicles

For ordinary (Fickian) Brownian motion, the mean-squared displacement (MSD) grows linearly with time: ⟨Δx²(t)⟩ = 2dDt (d = number of dimensions). Single-particle tracking of vesicles, mRNA granules and large protein complexes inside living cells instead shows:

This is anomalous sub-diffusion. Three overlapping mechanisms are usually invoked:

Sub-diffusion is not a violation of the fluctuation-dissipation theorem — it reflects a memory kernel in the friction (a "generalised Langevin equation" with time-correlated noise) rather than the instantaneous, memoryless drag assumed in the simple Stokes-Einstein picture.

5. Active Transport vs. Passive Jitter

Passive diffusion of a 500 nm vesicle across a 20 µm cell, using D ≈ 0.01-0.1 µm²/s, would take from tens of minutes to many hours — far too slow for cellular logistics. This is why eukaryotic cells evolved motor-protein-driven transport: kinesin and dynein walk cargo along microtubule tracks at 0.5-2 µm/s, directed and ballistic rather than random, covering the same distance in seconds to minutes. Single-particle tracking data typically show vesicles alternating between diffusive runs (α < 1, as in §4) and fast, directed, motor-driven runs (effectively α ≈ 2, ballistic) — a signature used experimentally to distinguish free cargo from motor-engaged cargo.

6. JavaScript: A Crowded-Cytoplasm Langevin Simulation

The simplest way to reproduce sub-diffusion numerically is a fractional or "obstructed" Langevin integrator: an ordinary overdamped Langevin step, but with a time-varying local diffusion coefficient sampled from a spatial obstacle map, mimicking the crowded mesh of §3.

class CrowdedCytoplasmWalker {
  constructor(D0 = 0.05, gridSize = 64, obstacleFraction = 0.28) {
    this.D0 = D0;             // free-space diffusion coefficient, µm²/s
    this.x = 0; this.y = 0;
    this.dt = 0.001;        // s
    this.grid = buildObstacleGrid(gridSize, obstacleFraction);
  }

  // local diffusivity drops sharply near obstacles (crowding)
  _localD(x, y) {
    const occ = occupancyNear(this.grid, x, y);
    return this.D0 * Math.max(0.02, 1 - occ);
  }

  step() {
    const D = this._localD(this.x, this.y);
    const sigma = Math.sqrt(2 * D * this.dt);
    this.x += sigma * gaussianRandom();
    this.y += sigma * gaussianRandom();
    return { x: this.x, y: this.y };
  }
}

// Run 10,000 steps and fit MSD(t) ~ t^alpha to recover the sub-diffusion exponent
const walker = new CrowdedCytoplasmWalker();
const trace  = [];
for (let i = 0; i < 10000; i++) trace.push(walker.step());
// log-log regression of ⟨Δx(t)²⟩ against t typically returns alpha ≈ 0.6-0.8
// for obstacleFraction = 0.28, matching measured vesicle tracks

Fitting the resulting MSD curve on log-log axes recovers an exponent α below 1, reproducing the qualitative anomalous-diffusion signature measured experimentally — without needing a full molecular-dynamics model of the crowded cytoplasm. For the derivation of why Gaussian random increments with variance 2Dt reproduce thermal noise in the first place, see the Brownian Motion & Langevin Equation article.