Noise-Driven Switching In Bistable Signaling

From Bistable States to Stochastic Switching

A deterministic bistable pathway at fixed conditions has two stable states, often described as low and high signaling activity, separated by an unstable state or basin boundary. In a deterministic model, a trajectory starting in one stable state stays in its basin and approaches that state. In a cell, however, molecular fluctuations can occasionally move the system out of that basin. The pathway may then settle near the other stable state even though the external conditions have not changed.

A useful coarse-grained description keeps only the two long-lived states and represents transitions between them as random events. It does not track the detailed signaling trajectory during a transition. This approximation is most plausible when the pathway spends much more time near either stable state than it spends crossing between them, and when the conditions governing the transition rates remain approximately constant over the time window of interest.

The Markov assumption means that, given the current state, the future transition probabilities do not depend on how long the pathway has already occupied that state. With constant rates, this implies exponentially distributed residence durations. These are model assumptions, not consequences of observing two clusters in a signaling measurement.

Residence Times and Switching Probabilities

While the pathway is in \(L\), its only possible exit in this model is a transition to \(H\). Let \(T_L\) be the duration of one uninterrupted visit to \(L\). Over a short interval, the probability of no exit is approximately \(1-k_{LH}\Delta t\). The memoryless, constant-rate model therefore gives an exponential survival probability. The corresponding result for a visit to \(H\) follows by replacing \(k_{LH}\) with \(k_{HL}\).

For a positive exit rate, the mean residence time is the reciprocal of that rate. If the exit rate is zero, the state is absorbing: the pathway never leaves it under this model, and the expected residence time is infinite in the extended sense.

Starting in \(L\), the probability of at least one transition to \(H\) by time \(t\) is one minus the probability of no exit from \(L\). This is a probability of having switched during the interval, not necessarily a probability of being in \(H\) at its endpoint: after switching, the pathway could switch back.

Occupancy at a Fixed Time

A different question is the probability of occupying \(H\) at a particular time. Write \(p_H(t)=P(X(t)=H)\). During a short interval, occupancy of \(H\) increases through transitions from \(L\) and decreases through transitions from \(H\). The forward equation is therefore

Set \(q=k_{LH}+k_{HL}\). If \(q>0\), the constant equilibrium solution of this equation is \(\pi_H=k_{LH}/q\). Subtracting this equilibrium from both sides gives \(d(p_H-\pi_H)/dt=-q(p_H-\pi_H)\). Solving this first-order equation with initial value \(p_H(0)\) yields the occupancy at time \(t\).

When \(q>0\), the stationary occupancy is unique, and the occupancy probability approaches it as time increases. If both rates are zero, neither state changes: \(p_H(t)=p_H(0)\) for every \(t\), so any initial occupancy is stationary and there is no unique stationary occupancy. This includes a pathway fixed in either state as well as an ensemble initially divided between states.

For example, if \(k_{LH}=0\) but \(k_{HL}>0\), \(H\) is not entered from \(L\), while a pathway already in \(H\) can leave it. The stationary occupancy is then \(\pi_H=0\). If \(k_{HL}=0\) but \(k_{LH}>0\), \(H\) is absorbing and the stationary occupancy is \(\pi_H=1\). These cases are well-defined because the rate sum is positive.

Noise-Driven Switching Versus Deterministic Crossing

In the two-state approximation, a transition can occur at fixed external conditions because the transition rates represent the effects of fluctuations on escape from one basin to the other. If \(k_{LH}>0\), the model assigns a nonzero probability of leaving \(L\) in any sufficiently long finite interval, even when the deterministic system would remain in the low-state basin. This is the meaning of noise-driven switching in this reduced description.

A deterministic threshold crossing is different. It occurs when a changing input or parameter moves the system’s state or changes the deterministic dynamics so that the trajectory crosses a boundary or loses the stability of its current state. It does not require a random transition event at fixed conditions. In data, crossing a chosen measurement threshold is not by itself proof that the system crossed a dynamical basin boundary: the threshold may simply classify measured activity as “low” or “high.”

A stochastic transition rate should be understood as a coarse-grained quantity for specified conditions. If the input changes substantially during the observation window, constant-rate formulas may not be appropriate. If rates vary with time but the two-state Markov description remains valid, the probability of no exit from a state with time-varying exit rate \(k(t)\) is \(e^{-\int_0^t k(u)\,du}\). For a pathway that is initially in \(L\), this describes survival in that initial visit, not its occupancy after possible returns.

Assumptions, Diagnostics, and Limitations

Before interpreting calculated probabilities, check whether the two-state approximation is credible for the question being asked. Its main assumptions are that the two states adequately represent the relevant long-lived signaling regimes, transitions are well-defined, and each state has an approximately constant exit rate over the interval analyzed. The memoryless assumption additionally requires that the chance of leaving soon does not depend on the duration already spent in that state.

• Inspect residence durations. Strong systematic dependence of exit probability on elapsed residence time, or a residence-time distribution that is markedly unlike an exponential, challenges the constant-rate Markov assumption.
• Check for unrepresented states. A reproducible intermediate signaling regime or multiple distinct transition routes can make a two-state model misleading.
• Check rate stability. If transitions accelerate or slow as input, drug concentration, or cell condition changes, a single constant rate may average over different regimes.
• Separate switching from measurement noise. Brief threshold crossings in a noisy readout may not represent a sustained transition between pathway states. State classification and time resolution affect observed switching counts.

A deterministic ordinary differential equation is a useful alternative when the question concerns trajectories and stability under specified conditions, but it does not describe random switching between stable states unless stochasticity is added. A detailed stochastic model can represent molecular fluctuations and transition paths more explicitly, but requires more states or variables and more information. The two-state continuous-time Markov model trades that mechanistic detail for compact, interpretable rates and probabilities. Its rates do not identify the molecular mechanism of switching by themselves, and its predicted probabilities should not be treated as evidence that a particular noise source caused a transition.