Why Positive Feedback Can Preserve a Signaling State
In a pathway with positive feedback, an active signaling component helps generate more of its own activity. A small initial signal can therefore be reinforced rather than simply transmitted and removed. If reinforcement is sufficiently nonlinear, the pathway may settle into either a low-activity or a high-activity state, depending on its initial condition and input history. This is bistable signaling: two distinct stable steady states coexist under the same fixed parameter values.
Bistability is not the same as a large response, a steep response, or a transient overshoot. It is a property of steady states and their stability. In a one-variable model, two stable equilibria are generally separated by an unstable equilibrium: small perturbations near either stable state decay, while perturbations near the unstable state grow away from it.
A Minimal Self-Activation Model
Let \(x\) denote normalized signaling activity and let \(\tau\) denote time measured in units of the activity-loss timescale. Consider a basal activation input, a self-activation term with a Hill response, and first-order loss:
Here \(b\geq 0\) is basal input, \(a\geq 0\) is the maximum feedback contribution, and \(n\) is the Hill coefficient. The Hill term is increasing and saturates at one; for \(n>1\), its response is especially weak near \(x=0\) and stronger over an intermediate activity range. The final term represents linear loss. The normalization sets the half-activation activity and the loss-rate scale to one. This is a simplified motif model, not a claim that all signaling components have these kinetics.
An equilibrium satisfies \(F(x^*)=0\). Graphically, it is an intersection of the activation rate \(b+a x^n/(1+x^n)\) and the loss rate \(x\). Whether an intersection is stable depends on how the rates change around it. For a one-dimensional differentiable model, a simple equilibrium is locally stable when \(F'(x^*)<0\), and unstable when \(F'(x^*)>0\).
Why the Slope Determines Local Stability
The stability rule follows by considering a small deviation \(\delta=x-x^*\) from an equilibrium. A first-order expansion of the rate gives \(F(x^*+\delta)\approx F(x^*)+F'(x^*)\delta\). Since \(F(x^*)=0\), the local deviation approximately obeys a linear differential equation.
A negative derivative makes the deviation shrink, whereas a positive derivative makes it grow. If \(F'(x^*)=0\), this linear test is inconclusive; behavior then depends on higher-order terms. In the examples below, the equilibria used for stability classification have nonzero derivatives.
Worked Example: Locating Two Stable States and the State Between Them
Suppose the normalized model has \(n=2\), \(a=2.25\), and \(b=0.0500\). Thus \(F(x)=0.0500+2.25x^2/(1+x^2)-x\). Find the equilibria and classify their local stability.
At equilibrium, multiply \(F(x)=0\) by \(1+x^2\), which is positive for all \(x\), to obtain
The cubic factors as \((x-0.5000)(x^2-1.800x+0.1000)=0\). Its three positive roots are \(x^*=0.5000\) and
giving \(x^*\approx0.05739\), \(0.5000\), and \(1.7426\), in normalized activity units. For \(n=2\), \(F'(x)=4.50x/(1+x^2)^2-1\). At the low equilibrium, \(F'(0.05739)\approx4.50(0.05739)/(1+0.05739^2)^2-1\approx-0.7435\), so it is stable. At the middle equilibrium, \(F'(0.5000)=2.25/(1.25)^2-1=0.4400\), so it is unstable. At the high equilibrium, \(F'(1.7426)\approx4.50(1.7426)/(1+1.7426^2)^2-1\approx-0.5188\), so it is stable.
Thus the same fixed basal input and feedback strength support two stable activity levels, separated by an unstable level. In this deterministic model, an initial activity below the unstable equilibrium tends toward the low state, while one above it tends toward the high state.
Feedback Strength and Basal Input
Positive feedback alone does not guarantee bistability. The feedback must be strong and nonlinear enough to make the net rate \(F(x)\) change direction so that it crosses zero three times. In the Hill model, the feedback contribution to the slope is \(a n x^{n-1}/(1+x^n)^2\). If this contribution stays below the loss slope of one, then \(F'(x)<0\) everywhere and \(F\) decreases strictly: there can be at most one equilibrium. Conversely, a region where the feedback slope exceeds one permits \(F\) to rise locally, making multiple intersections possible, though it does not by itself guarantee three equilibria for every value of \(b\).
For \(n=2\), the feedback slope is \(2ax/(1+x^2)^2\), whose maximum over \(x\geq0\) is \(3\sqrt{3}a/8\). For example, if \(a=1\), this maximum is \(3\sqrt{3}/8\approx0.6495<1\), so \(F'(x)<0\) throughout. With \(b=0.0500\), there is one equilibrium. Numerically, it is approximately \(x^*=0.05278\): using \(0.05278^2=0.0027857284\),
The small residual is consistent with a rounded numerical root. At this root, \(F'(0.05278)\approx2(0.05278)/(1+0.05278^2)^2-1\approx-0.8950\), confirming local stability. The slope argument establishes uniqueness; the numerical calculation illustrates its location.
Worked Example: Increasing Basal Input While Bistability Persists
Keep \(n=2\) and \(a=2.25\), but increase the basal input from \(b=0.0500\) to \(b=0.1000\). Determine whether there are still three equilibria, without assuming that their locations remain unchanged.
Now \(F(x)=0.1000+2.25x^2/(1+x^2)-x\). Evaluate the rate at four activities:
Because \(F\) is continuous, its sign changes imply at least one equilibrium in each of \((0,0.2000)\), \((0.2000,0.5000)\), and \((0.5000,2.000)\). Multiplying the equilibrium equation by \(1+x^2\) gives a cubic, so there can be no more than three equilibria. There are therefore exactly three distinct equilibria. The sign changes alternate from positive to negative, negative to positive, then positive to negative; the two outer crossings are stable and the middle crossing is unstable. Increasing basal input has not removed bistability in this case, although it changes the equilibrium locations.
This sign-check method establishes the number and ordering of equilibria here without claiming a universal basal-input threshold. Locating parameter values at which equilibria appear or disappear requires a separate analysis.
State Switching, Diagnostics, and Limits
With two stable states under the same fixed conditions, a sufficiently large transient input can move activity across the unstable equilibrium. Once the transient ends, the state may approach the other stable equilibrium. This is a state switch, not merely a larger temporary response. A change in basal input can also shift the equilibrium locations and, depending on the parameter range, alter whether the system remains bistable.
In simulations, inspect both the time course and the steady-state rate. A trajectory that appears to settle may only be changing slowly; verify that \(F(x)\) is close to zero and that the final state is not an artifact of the simulation duration or solver settings. To investigate coexistence, simulate from initial activities on both sides of the candidate unstable equilibrium, using identical fixed parameters. In experiments, compare steady outcomes after different input histories and allow enough time for transients to decay. Distinct outcomes are suggestive, but they do not alone prove that this specific feedback mechanism caused them.
A steep response curve is not sufficient evidence of bistability. A system can be ultrasensitive yet have only one stable equilibrium for each fixed input. Conversely, observing two steady outcomes under the same conditions requires ruling out explanations such as unmeasured differences between samples, slow adaptation, or persistent external inputs. The one-variable model also omits molecular detail, cell-to-cell variation, and stochastic fluctuations; its stable states are deterministic predictions, not guarantees that individual cells remain in a state indefinitely.
A common mistake is to say, “The feedback is positive, so the pathway must have two stable states.” Positive feedback can amplify activity, but bistability requires suitable nonlinearity and parameter values. Another mistake is to infer bistability from a steep dose-response curve alone. The correction is to identify equilibria under fixed conditions and assess their stability, rather than treating steepness or transient amplification as proof of multiple stable states.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- In the one-variable model, what sign of \(F'(x^*)\) indicates a locally stable equilibrium, and why?
- For \(n=2\), \(a=1\), and \(b=0.0500\), why can the model not have three equilibria?
- For the worked model with \(a=2.25\), \(b=0.0500\), and \(n=2\), which equilibrium separates the two stable states?
- Does a steep response curve by itself establish bistability? Give the reason.
Solutions. 1. \(F'(x^*)<0\): a small deviation approximately changes as \(\delta(\tau)=\delta(0)e^{F'(x^*)\tau}\), so it decays. 2. The maximum feedback slope is \(3\sqrt{3}/8\approx0.6495<1\), so \(F'(x)<0\) for all \(x\geq0\); the net rate is strictly decreasing and has at most one zero. 3. The middle equilibrium, \(x^*=0.5000\), is unstable and separates the low and high stable equilibria. 4. No. Steepness describes how rapidly a response changes, whereas bistability requires multiple steady states under the same fixed conditions and a stability assessment for each.