Modeling Directed Cell Migration by Chemotaxis

From a Chemical Gradient to Directed Migration

Chemotaxis is directed cell movement in response to a spatial gradient of a chemical signal. In a tissue model, the signal may be a ligand, cytokine, or other cue whose concentration varies with position. A useful model must say not only whether cells tend to move toward higher concentration, but also how that tendency combines with random motility, how it depends on gradient strength, and at what scale the movement rule is defined.

Two descriptions are common. A continuum model represents cells by a density field and summarizes migration as a flux. An agent-based model represents individual cells and specifies movement probabilities or rates. The descriptions can be related under suitable assumptions, but they are not interchangeable in every setting: a continuum approximation smooths individual trajectories and local heterogeneity, while an agent rule makes cell-level decisions explicit.

Definition: Let $c(\mathbf{x},t)$ be the concentration of a chemoattractant and $n(\mathbf{x},t)$ the cell density. A standard chemotactic flux is $$ \mathbf{J}=-D\nabla n+\chi n\nabla c. $$ Here $D>0$ is the random-motility coefficient and $\chi$ is the chemotactic sensitivity. For a chemoattractant, $\chi>0$; the directed component of the flux points up the concentration gradient. If $\chi<0$, it points down the gradient, as for a chemorepellent.

The units make the terms interpretable. If distance is measured in millimetres and time in hours, then $D$ has units mm²/h. If concentration is measured in ng/mL, $\chi$ has units mm²/(ng/mL)/h. The flux has units of cells crossing a unit area per unit time. Cell abundance changes according to conservation: in the absence of sources or sinks, $\partial n/\partial t=-\nabla\cdot\mathbf{J}$. Proliferation, death, or other processes can be added as separate terms; they are not part of chemotactic movement itself.

Interpreting the Continuum Flux

The diffusive term $-D\nabla n$ moves cells down their own density gradient. It tends to smooth spatial differences in cell density. The chemotactic term $\chi n\nabla c$ is different: it moves cells in response to the chemical gradient, with a magnitude proportional to the local density $n$. If there are no cells at a location, this flux formula gives no cell flux there, regardless of the signal.

The flux is not the same quantity as the local rate of change in cell density. A positive flux in the $x$-direction means cells move across a surface toward increasing $x$. Whether density rises at a particular location depends on the divergence of the flux: more cells entering a small region than leaving it increases its density. In particular, a spatially uniform cell density can still have a nonzero chemotactic flux if the chemical gradient is nonzero.

The simplest version assumes $D$ and $\chi$ are constant and that cells respond instantaneously to the local gradient. In general, $\chi$ may depend on concentration or cell state. The conservation equation must then retain the full divergence $-\nabla\cdot(\chi n\nabla c)$; it is not generally valid to treat $\chi$ as constant when differentiating. A model also needs spatial boundary conditions. For an impermeable tissue boundary, the normal component of total cell flux is zero there, unless the biology permits cells to enter or leave.

Worked Example: Combining Random and Directed Fluxes

Consider a hypothetical two-dimensional tissue region. At one location, let the cell density be $100$ cells/mm², with gradient $\nabla n=(10,0)$ cells/mm³. Let the chemoattractant gradient be $\nabla c=(0.20,0)$ (ng/mL)/mm, with $D=0.080$ mm²/h and $\chi=0.30$ mm²/(ng/mL)/h. Assume these local values are adequate to evaluate the flux and that no source or sink term is being included.

The diffusive component is $$ -D\nabla n=-(0.080)(10,0)=(-0.80,0)\ \text{cells}/(\text{mm}\cdot\text{h}). $$ The chemotactic component is $$ \chi n\nabla c=(0.30)(100)(0.20,0)=(6.00,0)\ \text{cells}/(\text{mm}\cdot\text{h}). $$ Thus the total local flux is $$ \mathbf{J}=(-0.80,0)+(6.00,0)=(5.20,0)\ \text{cells}/(\text{mm}\cdot\text{h}). $$

The net flux points in the positive $x$-direction: here the directed contribution exceeds the opposing diffusive contribution. This calculation alone does not establish whether density at the location is increasing. That requires the spatial divergence of the flux, which describes the local balance of inflow and outflow.

From Individual Jumps to Drift and Diffusion

An agent-level rule can encode gradient sensing by assigning a higher probability or rate to movement toward greater concentration. For a simple one-dimensional lattice, let adjacent sites be separated by $h$, and let an agent make a left or right jump at attempt rate $\lambda$. Write $\Delta c=c(x+h)-c(x-h)$ for the concentration difference used by the cell to compare the two directions. One smooth choice is $$ p_{\mathrm{right}}=\frac{\exp(\beta\Delta c)}{\exp(\beta\Delta c)+\exp(-\beta\Delta c)},\qquad p_{\mathrm{left}}=1-p_{\mathrm{right}}, $$ where $\beta$ has units reciprocal to concentration. For $\beta>0$, the probability of moving right exceeds one half when the rightward comparison indicates higher concentration. This rule assumes that a cell senses the signal over the specified comparison distance and that its direction choice is described by these probabilities.

The signed displacement $\xi$ in one attempt is $+h$ with probability $p_{\mathrm{right}}$ and $-h$ with probability $p_{\mathrm{left}}$. Therefore, $$ E[\xi]=h(p_{\mathrm{right}}-p_{\mathrm{left}})=h\tanh(\beta\Delta c). $$ This is the expected signed displacement per attempt, with units of length per attempt. If attempts occur at rate $\lambda$, the mean drift speed is $\lambda E[\xi]$, with units of length per time. Keeping those quantities distinct prevents an important dimensional error.

For a small lattice spacing and a smooth signal, $\Delta c$ is approximately $2h\,\partial c/\partial x$. When $|\beta\Delta c|$ is small, $\tanh(\beta\Delta c)$ is approximately $\beta\Delta c$, so the mean drift speed is approximately $2\lambda\beta h^2\,\partial c/\partial x$. The unbiased random-jump contribution has effective diffusion coefficient $D=\lambda h^2/2$. Thus, in this small-step, weak-bias limit, the corresponding continuum chemotactic sensitivity is $\chi\approx 2\lambda\beta h^2=4D\beta$ under this particular definition of $\Delta c$. If a rule instead compares concentrations across a single spacing, or defines its sensing parameter differently, the conversion factor changes. The movement rule and sensing distance must be stated before interpreting $\chi$.

Derivation: In a short time interval, a cell makes jumps with mean displacement per attempt $E[\xi]$ at rate $\lambda$. Its mean drift is therefore $a=\lambda E[\xi]$. The second raw jump moment is $E[\xi^2]=h^2$, so the second-moment coefficient is $b=\lambda h^2$. For small jumps and smoothly varying quantities, the one-dimensional density equation has the drift-diffusion form $$ \frac{\partial n}{\partial t}=-\frac{\partial (a n)}{\partial x}+\frac{1}{2}\frac{\partial^2 (b n)}{\partial x^2}. $$ When $b$ is constant, this is equivalent to a flux $J=a n-(b/2)\partial n/\partial x$: drift carries density at speed $a$, while diffusion has coefficient $D=b/2$. This local-moment approximation requires jump lengths to be small relative to the spatial scale on which density and signal vary.

Worked Example: Expected Displacement Is Not Speed

Suppose a hypothetical lattice model has spacing $h=20.0$ μm, attempt rate $\lambda=0.500$ attempts/min, and $\beta\Delta c=0.200$. The probabilities are $$ p_{\mathrm{right}}=\frac{1}{1+\exp(-0.400)}\approx0.5987,\qquad p_{\mathrm{left}}\approx0.4013. $$

The expected signed displacement per attempt is $$ E[\xi]=(20.0)(0.5987-0.4013)=3.948\ \text{μm per attempt}. $$ Multiplying by the attempt rate gives the mean drift speed: $$ v=\lambda E[\xi]=(0.500)(3.948)=1.974\ \text{μm/min}. $$

The units distinguish the results: $3.948$ μm is an expected displacement for one attempt, whereas $1.974$ μm/min is a mean displacement rate over time. With no bias, the corresponding lattice diffusion coefficient would be $D=\lambda h^2/2=(0.500)(20.0^2)/2=100$ μm²/min. This coefficient describes spread due to random jumps; it is not the directed speed.

How Stronger Gradients Change Migration

Under a fixed movement rule, increasing the sensed concentration difference increases directional bias. For the smooth probability rule above, the response is approximately linear when $|\beta\Delta c|$ is small, but it saturates as the difference becomes large: the probability of choosing the favored direction approaches one rather than growing without bound. In the continuum flux, a constant positive $\chi$ gives a chemotactic flux proportional to $\nabla c$, so that formula does not itself include saturation. A saturating agent rule may therefore require a concentration-dependent effective sensitivity or may differ from a linear continuum model outside the weak-gradient regime.

Worked Example: Comparing Two Gradient Strengths

Use the hypothetical lattice settings $h=20.0$ μm and $\lambda=0.500$ attempts/min from the previous calculation. Compare two local sensing differences, $\beta\Delta c=0.200$ and $\beta\Delta c=0.600$. These are dimensionless products, so the comparison changes signal contrast while keeping the movement rule and attempt rate fixed.

For $\beta\Delta c=0.200$, $p_{\mathrm{right}}=1/(1+\exp(-0.400))\approx0.5987$, giving $E[\xi]=20.0(2p_{\mathrm{right}}-1)\approx3.948$ μm per attempt and $v\approx(0.500)(3.948)=1.974$ μm/min.

For $\beta\Delta c=0.600$, $p_{\mathrm{right}}=1/(1+\exp(-1.200))\approx0.7685$. Then $$ E[\xi]=20.0(2(0.7685)-1)\approx10.74\ \text{μm per attempt},\qquad v=(0.500)(10.74)\approx5.370\ \text{μm/min}. $$

The stronger sensed difference yields greater mean drift toward the favored direction. It does not imply that every cell moves right, nor that cell density at every point increases. Density patterns also depend on where cells are, the spatial variation in flux, random motility, boundaries, and any source or sink processes.

Assumptions, Diagnostics, and Choice of Description

A continuum flux is useful when the cell population is sufficiently dense for a smooth density field to be meaningful, and when movement can be summarized by local drift and diffusion. Its coefficients should be interpreted at the spatial and temporal scales of the model. If individual cells make long jumps, respond after a delay, or retain direction over time, a local instantaneous drift-diffusion flux may not capture their behavior without additional state variables or terms.

An agent-level rule is useful when individual choices, cell-cell interactions, occupancy constraints, or sparse populations matter. Its output depends on implementation details such as update schedule, sensing distance, attempted-move frequency, and how occupied sites or boundaries are handled. A model that uses the same probabilities but a different attempt rate does not have the same mean speed. A lattice rule may also impose directional artifacts if its allowed movement directions do not represent the intended geometry.

Useful diagnostics include checking that the drift points toward higher $c$ for a chemoattractant, verifying the units of displacement and speed, and comparing observed mean displacement and spread with the rule's implied moments. In a spatial simulation, inspect cell density and flux near boundaries as well as in the interior. Strong accumulation at an impermeable boundary can be physically plausible, but unexpected loss of cells or a persistent boundary-directed flux can indicate a boundary-condition or implementation problem. Mesh and time-step checks are also important when a continuum equation is discretized; a pattern that changes substantially with refinement may be numerical rather than biological.

A common mistake is to interpret $h(p_{\mathrm{right}}-p_{\mathrm{left}})$ as a speed. That quantity is expected signed displacement per attempt, with units of length per attempt. The speed is $\lambda h(p_{\mathrm{right}}-p_{\mathrm{left}})$, with units of length per time. Another frequent error is to treat a directed flux as proof that local density rises: density change depends on the divergence of total flux, not flux direction alone.

Key takeaway: Chemotactic migration can be represented as a directed continuum flux or as biased individual movement. Interpret either model through its stated sensing rule, units, scale assumptions, and boundary handling; a stronger gradient generally increases directional bias, but does not by itself determine the full spatial density pattern.

Check Your Understanding

Answer each question, then compare with the brief solutions below.

If $D=0.10$ mm²/h, $n=50$ cells/mm², and $\partial c/\partial x=0.30$ (ng/mL)/mm with $\chi=0.20$ mm²/(ng/mL)/h, what is the chemotactic flux in the $x$-direction?
A lattice rule gives an expected signed displacement of $2.0$ μm per attempt and an attempt rate of $0.25$ attempts/min. What is the mean drift speed?
Why does a positive flux in the $x$-direction not, by itself, show that local cell density is increasing?
For the stated probability rule, what happens to $p_{\mathrm{right}}$ as a positive $\beta\Delta c$ becomes very large, and what does that imply about response saturation?

Solutions. (1) $J_{\mathrm{chem}}=\chi n\,\partial c/\partial x=(0.20)(50)(0.30)=3.00$ cells/(mm·h), directed toward increasing $x$. The given $D$ is not needed for the chemotactic component. (2) $v=(0.25)(2.0)=0.500$ μm/min; the displacement per attempt is not itself a speed. (3) Density change depends on $-\nabla\cdot\mathbf{J}$: local density rises when inflow exceeds outflow, not merely because flux points in one direction. (4) $p_{\mathrm{right}}=1/(1+\exp(-2\beta\Delta c))$ approaches one. The probability cannot exceed one, so directional bias saturates rather than increasing without bound.