Toggle Switches From Mutual Inhibition

Why Mutual Inhibition Can Make a Switch

Two signaling components that repress one another can support alternative states: one component is abundant while the other is suppressed, or the roles are reversed. This architecture is called a toggle switch. It is useful when a system must maintain one of two expression or activity patterns after the initiating signal changes. Mutual inhibition does not guarantee bistability, however. Whether the system has two stable states depends on the strengths and shapes of the inhibition responses.

We will use a deterministic, continuous-time model with Hill-function repression. Concentrations $x$ and $y$ are scaled to dimensionless units, and time $\tau$ is scaled so that each component has a first-order degradation rate of one. The model represents constant production drives, mutual repression, and first-order removal. It omits delays, intrinsic molecular noise, and other regulatory interactions.

Definition: A two-component mutual-inhibition model with squared Hill repression is $$ \frac{dx}{d\tau}=\frac{A}{1+\beta y^2}-x,\qquad \frac{dy}{d\tau}=\frac{B}{1+\gamma x^2}-y. $$ Here $A$ and $B$ are positive production-drive parameters, while $\beta$ measures inhibition of $x$ by $y$ and $\gamma$ measures inhibition of $y$ by $x$. All variables and parameters are dimensionless. An equilibrium is a state where both derivatives are zero; it is locally asymptotically stable if nearby states approach it over time.

The squared Hill terms are a phenomenological representation of cooperative, switch-like repression. They are not a claim that every molecular interaction has precisely this form. The model assumes that each production rate responds instantaneously to the other component and that removal is proportional to the component's concentration. These assumptions make the nullclines and stability conditions explicit.

Nullclines and the Stability Test

A nullcline is the set of states where one component's rate of change is zero. Equilibria occur where the two nullclines intersect:

x=\frac{A}{1+\beta y^2},\qquad y=\frac{B}{1+\gamma x^2}.

Each nullcline slopes downward in the $x,y$ plane: increasing one component reduces the production of the other. Multiple intersections are possible when repression is sufficiently strong and nonlinear. To determine whether an intersection is stable, linearize the model around it. At an equilibrium, the Jacobian matrix is

J= \begin{pmatrix} -1 & -\dfrac{2\beta xy}{1+\beta y^2}\\[6pt] -\dfrac{2\gamma xy}{1+\gamma x^2} & -1 \end{pmatrix}.

The equilibrium substitutions in the off-diagonal entries follow from $A=x(1+\beta y^2)$ and $B=y(1+\gamma x^2)$. The trace is $-2$, and the determinant is

\det(J)=1-G,\qquad G=\frac{4\beta\gamma x^2y^2} {(1+\beta y^2)(1+\gamma x^2)}.

For a two-dimensional continuous-time system, an equilibrium is locally asymptotically stable when the trace is negative and the determinant is positive. Here the trace is always negative, so $G<1$ implies stability. If $G>1$, the determinant is negative and the equilibrium is a saddle. The boundary $G=1$ is the local condition for a saddle-node fold, provided the parameter change crosses the fold nondegenerately.

Worked Example: Finding Both Stable States

Take the symmetric model with $A=B=4$ and $\beta=\gamma=1$. Find its equilibria and classify them. The nullclines are $x=4/(1+y^2)$ and $y=4/(1+x^2)$. Multiplying each equation by its denominator gives $x(1+y^2)=4$ and $y(1+x^2)=4$.

Subtracting the equations and factoring gives

$$ x(1+y^2)-y(1+x^2)=(x-y)(1-xy)=0. $$

Thus either $x=y$, or $xy=1$. On the diagonal $x=y$, the equilibrium solves $x^3+x-4=0$, with positive root $x\approx1.379$. At that point, $G=(2x^2/(1+x^2))^2\approx1.718$, so $\det(J)\approx-0.718$: the diagonal equilibrium is a saddle.

For the other two equilibria, $xy=1$, and the first equilibrium equation becomes $x+1/x=4$. Therefore $x=2\pm\sqrt{3}$, with $y=1/x$. The states are approximately $(3.732,0.2679)$ and $(0.2679,3.732)$. At either one, $G=4/[(1+x^2)(1+y^2)]=4/16=0.25$, so $\det(J)=0.75>0$. Both are stable. The model therefore has two alternative stable states separated by a saddle.

From a Fold Condition to a Switching Threshold

A switching threshold is a parameter value at which a stable equilibrium and a saddle meet and disappear. At that point, one eigenvalue of the Jacobian is zero, so its determinant is zero. This gives $G=1$. The condition can be combined with the equilibrium equations to calculate the fold in terms of the model parameters.

For a fixed pair of inhibition strengths, a convenient way to describe the fold is to choose the concentrations $x,y$ at the fold and calculate the corresponding drives:

A=x(1+\beta y^2),\qquad B=y(1+\gamma x^2),\qquad \frac{4\beta\gamma x^2y^2}{(1+\beta y^2)(1+\gamma x^2)}=1.

These equations locate candidate saddle-node points. A candidate should also be checked to ensure the fold is nondegenerate and lies on the parameter path of interest. As a parameter is swept, a system with two stable branches can remain on one branch until its fold is reached; it can then move toward the other stable state. The two fold values need not coincide, which is the basis of hysteresis described in the earlier material on saddle-node bifurcations.

Worked Example: Calculating a Fold Under Asymmetric Inhibition

Set $\beta=1$ and $\gamma=4$, so inhibition of $y$ by $x$ is stronger than inhibition of $x$ by $y$. Hold $B=32/11$ fixed and find a fold on this parameter slice. Define $t=\gamma x^2$. The fold condition becomes

\frac{4ty^2}{(1+y^2)(1+t)}=1 \quad\Longrightarrow\quad (3t-1)y^2=1+t.

Since $B=y(1+t)$, this also implies $B^2=(1+t)^3/(3t-1)$. The value $t=5/11$ satisfies this equation for $B=32/11$, because

\frac{(1+5/11)^3}{3(5/11)-1} =\frac{(16/11)^3}{4/11} =\frac{1024}{121} =\left(\frac{32}{11}\right)^2.

Consequently, $y=B/(1+t)=2$, and $x=\sqrt{t/\gamma}=\sqrt{5/44}\approx0.3371$. The fold drive for $x$ is

A=x(1+y^2)=5\sqrt{\frac{5}{44}}\approx1.685.

For comparison, keep $\beta=1$ and $B=32/11$ but change $\gamma$ to 1. The same fold equation for $t=x^2$ has the solution $t=5/11$, giving $y=2$, $x=\sqrt{5/11}$, and $A=5\sqrt{5/11}\approx3.371$. On this fold branch, increasing $\gamma$ from 1 to 4 therefore lowers the $x$-production threshold from about 3.371 to 1.685, with $B$ and $\beta$ held fixed. The numerical shift is conditional on this model and parameter slice; asymmetric inhibition does not imply a universal direction of threshold change for every parameter change.

How to Read the Switch, and Where the Model Can Fail

The important evidence for a toggle is not simply that both nullclines slope downward. It is that they intersect in three equilibria, with two stable states and an intervening saddle. The example with $A=B=4$ demonstrated this pattern. If parameters are changed, the nullcline intersections and their stability can change; a single intersection is not a bistable switch. Numerical root finding can help locate intersections, but the Jacobian test is still needed to classify them.

In practice, a useful analysis proceeds from the model to its equilibria and then to the parameter-dependent folds. For a parameter sweep, continuation methods can trace equilibrium branches and help detect folds that a coarse grid of simulations might miss. Simulations from different initial conditions can reveal convergence to different stable states, but they do not by themselves establish where a fold lies or prove that no other equilibria exist.

State the model and scaling.
Specify the production drives, inhibition strengths, degradation assumptions, and the units or dimensionless scales used.

Find nullcline intersections.
Solve the two equilibrium equations, checking that candidate concentrations are nonnegative.

Classify each intersection.
Evaluate $G$. Values below one give stable equilibria; values above one give saddles for this model.

Locate and interpret folds.
Use $G=1$ together with the equilibrium equations, then check that the fold is nondegenerate and relevant to the parameter sweep.

Mutual inhibition is often compared with a model in which each component activates itself. Both architectures can generate alternative stable states, but their regulatory interpretations differ: mutual inhibition suppresses the competing state, whereas autoactivation reinforces a component's own activity. A Hill-function toggle is simpler to analyze than a detailed binding and turnover model, but its parameters are effective quantities and may not correspond directly to individual molecular rate constants.

Common failure modes include treating a steep response as proof of bistability, calling every nullcline intersection stable, and interpreting a fold value as a universal switching threshold. A steep but single-valued response can have only one equilibrium. In this model, stability requires checking the determinant, not just inspecting the nullcline slopes. A threshold depends on the other parameters and on which parameter is being varied. Finally, deterministic equations describe stable states and transitions under changing parameters, but do not describe random switching caused by molecular fluctuations. If fluctuations or delays are important, this model's assumptions are insufficient.

Key takeaway: In the squared-Hill mutual-inhibition model, alternative stable states are identified by multiple nullcline intersections and the Jacobian test. The fold condition $G=1$ gives a way to calculate switching thresholds, and unequal inhibition strengths can shift those thresholds substantially.

Check Your Understanding

Try each question, then check the solutions below.

For the stated model, what do $\beta$ and $\gamma$ represent?
At an equilibrium with $G=0.6$, classify the equilibrium using the trace and determinant.
At a point where $G=1$, what additional feature is needed before calling it a nondegenerate saddle-node fold?
Why is a steep response curve alone insufficient evidence for a toggle switch?

Solutions. (1) $\beta$ measures inhibition of $x$ by $y$, and $\gamma$ measures inhibition of $y$ by $x$. (2) The trace is $-2$ and the determinant is $1-0.6=0.4>0$, so the equilibrium is locally asymptotically stable. (3) The parameter change must cross the fold nondegenerately; the equilibrium equations and fold condition alone identify a candidate that should be checked. (4) A steep response can still have a single equilibrium. A toggle requires alternative stable equilibria, verified by solving for intersections and checking their stability.

Toggle Switches From Mutual Inhibition

What You'll Learn