How Signaling Cascades Encode Pulsatile Inputs

Why Pulse Timing Matters

A pathway does not necessarily reproduce every change in its input. Each signaling layer takes time to respond, so a brief input may be attenuated, delayed, or spread over time before it reaches a downstream target. Repeated inputs can also arrive before the pathway has returned to baseline. In that case, residual activity from one pulse contributes to the response to the next.

A useful starting point is a cascade of first-order signaling steps. It makes the effects of layer number, time constants, and pulse timing calculable. The model is an idealization: it describes linear filtering, not every biochemical detail of phosphorylation, saturation, or feedback.

Definition: In a normalized first-order cascade with $n$ layers, the input $u(t)$ drives the first layer, and each layer $x_i(t)$ relaxes toward the activity of the preceding layer: $$ \frac{dx_i}{dt}=\frac{x_{i-1}(t)-x_i(t)}{\tau_i},\qquad i=1,\ldots,n,\qquad x_0(t)=u(t). $$ Here $x_i$ is the activity in layer $i$, and $\tau_i>0$ is its time constant, in units of time. The output is $x_n(t)$. The normalized model assumes unit steady-state gain at every layer; a constant input is eventually reproduced at every layer.

This model assumes that each layer has linear, time-invariant dynamics; the time constants and gains do not change during the input; and the pathway begins in the stated initial condition. In the calculations below, the initial activity is zero, the baseline input is zero, and pulses are rectangular. These assumptions allow superposition: the response to a sum of inputs is the sum of the responses to each input. They are not appropriate without qualification when the pathway saturates or has state-dependent kinetics.

Response to One Pulse

First consider identical time constants, $\tau_i=\tau$. A step input of amplitude $A$, switched on at time zero, produces a progressively delayed response as it passes through the cascade. Define $F_n(t)$ as the output divided by $A$ after a unit step has been applied for $t$ time units. For $t\geq 0$,

F_n(t)=1-e^{-t/\tau}\sum_{k=0}^{n-1}\frac{(t/\tau)^k}{k!}.

For $t<0$, define $F_n(t)=0$. A rectangular pulse of amplitude $A$ and duration $d$ is the difference between a step switched on at time zero and an equal step switched on at time $d$. Therefore its output is

x_n(t)=A\left[F_n(t)-F_n(t-d)\right].

The subtraction matters after the pulse ends: it represents the downstream response to the input switching off. The output does not generally drop to zero at that moment, because activity already in the cascade continues to progress through its layers.

Justification: For one layer, the step response is $F_1(t)=1-e^{-t/\tau}$, obtained by solving $\tau\,dx_1/dt=A-x_1$ from $x_1(0)=0$ and dividing by $A$. Each additional layer filters the preceding layer's response according to the same first-order equation. Equivalently, the cascade has impulse response $h_n(t)=e^{-t/\tau}t^{n-1}/(\tau^n(n-1)!)$ for $t\geq0$; integrating this response from 0 to $t$ gives $F_n(t)$. Repeated integration by parts yields $1-F_n(t)=e^{-t/\tau}\sum_{k=0}^{n-1}(t/\tau)^k/k!$, which gives the stated expression.

Worked Example: Comparing One, Two, and Three Layers

Suppose a hypothetical pathway receives a unit-amplitude input pulse lasting 4 minutes. The pathway has identical 2-minute time constants, starts at baseline, and has either one, two, or three normalized layers. What is each output at the instant the input pulse ends?

At $t=4$ minutes, the input has been on for $d=4$ minutes and has not yet been off for any time. Thus $x_n(4)=F_n(4)$, with $t/\tau=4/2=2$.

For one layer, $$ F_1(4)=1-e^{-2}=1-0.1353=0.8647. $$

For two layers, $$ F_2(4)=1-e^{-2}(1+2)=1-3(0.1353)=0.5940. $$

For three layers, $$ F_3(4)=1-e^{-2}\left(1+2+\frac{2^2}{2}\right) =1-5(0.1353)=0.3233. $$

At pulse offset, the outputs are about 0.865, 0.594, and 0.323 output units, respectively. In this model, adding layers makes the response slower: less of the input's change has reached the output by the end of the pulse. The calculation does not imply that real biochemical cascades always have unit gain or that adding a layer must reduce every possible measure of pathway activity.

For the two-layer cascade, two minutes after the input has switched off, $t=6$ minutes. The delayed-step formula gives $$ x_2(6)=F_2(6)-F_2(2) =\left[1-4e^{-3}\right]-\left[1-2e^{-1}\right] =0.8009-0.2642=0.5366. $$ The output remains positive after the input ends. It is lower than its value at pulse offset, but it has not returned to baseline.

Repeated Pulses and Residual Activity

For periodic pulses, the response depends on both the pulse duration and the time available for recovery between pulses. A one-layer cascade makes this dependence especially transparent. Let pulses of amplitude $A$ last $w$ time units and repeat every $T$ time units, where $T>w$. At periodic steady state, let $P$ be the output just as a pulse ends, and let $B$ be the output just before the next pulse begins. During a pulse the output approaches $A$; during the interval without input it decays toward zero.

P=A(1-e^{-w/\tau})+e^{-w/\tau}B,\qquad B=e^{-(T-w)/\tau}P.

Solving these two relations gives

P=\frac{A(1-e^{-w/\tau})}{1-e^{-T/\tau}}, \qquad B=e^{-(T-w)/\tau}P.

The denominator accounts for activity carried over from earlier cycles. If the off interval is long relative to $\tau$, little remains before the next pulse. If pulses arrive more frequently, the trough rises because the output has less time to decay. In a linear system, multiplying every pulse amplitude by a constant multiplies both the peak and trough by that same constant, without changing their ratio.

Worked Example: Short Versus Long Pulse Intervals

Consider a hypothetical one-layer pathway with $\tau=3$ minutes and pulses of amplitude 1 output unit lasting 1 minute. Calculate the periodic steady-state peak and trough for repetition periods of 4 and 10 minutes. The pulse width and amplitude are unchanged; assume the pathway has reached periodic steady state.

For the 4-minute period, the off interval is $4-1=3$ minutes: $$ P=\frac{1(1-e^{-1/3})}{1-e^{-4/3}} =\frac{1-0.7165}{1-0.2636} =\frac{0.2835}{0.7364}=0.3849. $$ The trough is $$ B=e^{-3/3}P=e^{-1}(0.3849)=0.1416. $$

For the 10-minute period, the off interval is $10-1=9$ minutes: $$ P=\frac{1(1-e^{-1/3})}{1-e^{-10/3}} =\frac{0.2835}{1-0.03567}=0.2939, \qquad B=e^{-9/3}(0.2939)=e^{-3}(0.2939)=0.01464. $$

The more frequent train has a higher peak and a much higher trough: approximately 0.385 and 0.142 units, compared with 0.294 and 0.0146 units. The output accumulates because the pathway retains activity between pulses. The time-averaged input is also different: amplitude times pulse width divided by period gives 0.250 units for the 4-minute period and 0.100 units for the 10-minute period.

Worked Example: Equal Average Input, Different Pulse Patterns

Compare two hypothetical periodic inputs to a one-layer pathway with $\tau=2$ minutes. Pattern A has amplitude 2 units, duration 1 minute, and period 6 minutes. Pattern B has amplitude 1 unit, duration 1 minute, and period 3 minutes. Both have an average input of $1/3$ unit: for A, $2(1)/6=1/3$; for B, $1(1)/3=1/3$. Calculate their steady-state peaks and troughs.

For both patterns, $e^{-w/\tau}=e^{-1/2}=0.6065$, so $1-e^{-w/\tau}=0.3935$. For A, the period is 6 minutes: $$ P_A=\frac{2(0.3935)}{1-e^{-6/2}} =\frac{0.7869}{1-e^{-3}} =\frac{0.7869}{0.9502}=0.8282. $$ The off interval is 5 minutes, so $$ B_A=e^{-5/2}(0.8282)=0.08208(0.8282)=0.06800. $$

For B, the period is 3 minutes: $$ P_B=\frac{1(0.3935)}{1-e^{-3/2}} =\frac{0.3935}{1-0.2231} =\frac{0.3935}{0.7769}=0.5065, $$ and $$ B_B=e^{-2/2}(0.5065)=e^{-1}(0.5065)=0.1863. $$

The two trains have the same average input, but their output peaks and troughs differ. Pattern A has the larger peak, about 0.828 units; Pattern B has the larger trough, about 0.186 units. Thus the average alone does not describe the temporal pattern reaching the output. A downstream response that depends nonlinearly on activity may distinguish these patterns, even when their average inputs match. This calculation does not specify that downstream response.

Interpreting Cascade Depth and Time Constants

For a finite pulse, the dimensionless ratio $d/\tau$ is a useful guide. A pulse much shorter than a layer's time constant produces only a partial response in that layer. A pulse much longer than the relevant time constants allows more of the pathway to approach its pulse-on level. Adding layers generally spreads and delays the response, so narrow, rapidly repeated inputs are less faithfully reproduced at the output than slow changes. The exact output still depends on the pulse duration, layer kinetics, gains, and the quantity being compared, such as peak, trough, or time average.

With unequal time constants, the simple $F_n$ expression above does not apply. The cascade equations still define the model: solve each first-order equation in order, using the preceding layer's time course as its input. For a cascade with fixed gains and time constants, its linear time-invariant representation has transfer function $$ H(s)=\prod_{i=1}^{n}\frac{1}{1+\tau_i s}, $$ where $s$ denotes the Laplace variable. If layer $i$ has a fixed gain $g_i$, the corresponding factor is $g_i/(1+\tau_i s)$, and the overall steady-state gain is $\prod_i g_i$. This representation makes clear that each layer contributes its own filtering; it does not make unequal layers identical.

The linear model also predicts that scaling pulse amplitude scales the entire output by the same factor. If phosphorylation sites approach saturation, thresholds are crossed, or feedback changes the effective kinetics, this proportionality can fail. The earlier material on ultrasensitivity and feedback is relevant to recognizing those mechanisms, but the calculations here do not model them. In such settings, use the appropriate nonlinear pathway equations rather than interpreting the linear-filter result as a universal rule.

Diagnostics, Alternatives, and Limits

Check the setup before interpreting a pulse calculation. Confirm that the time units for pulse width, period, and time constants agree; that $T>w$ for separated periodic pulses; and that a repeated-pulse calculation is being interpreted at periodic steady state rather than during the first cycle. For an isolated pulse, the off-step subtraction must be included whenever $t\geq d$. Omitting it incorrectly treats the input as if it remained on forever.

A simple diagnostic is to examine limiting cases. In the one-layer periodic result, a very long off interval makes $e^{-(T-w)/\tau}$ small, so the trough approaches zero and the peak approaches the response to one isolated pulse. As the period approaches the pulse width from above, there is almost no off interval, so the trough approaches the peak. If a calculation violates these expectations, check the timing convention or algebra.

The alternatives depend on what is known about the pathway. For identical first-order layers, the closed-form step response is efficient and exposes how layer count affects pulse shape. For unequal time constants or complex input histories, numerical integration of the cascade equations is flexible and can directly produce the requested time course. A nonlinear biochemical model may be needed if gain depends on signal level or pathway state. These approaches answer related but not interchangeable questions: a linear cascade isolates filtering, while a detailed nonlinear model can represent mechanisms omitted by that idealization.

A cascade output can preserve information about amplitude and timing, but filtering can make distinct input patterns look similar. The model alone does not establish that a cell decodes pulse frequency, that a particular output causes a phenotype, or that a measured signal corresponds exactly to $x_n$. Such claims require an appropriate measurement model and biological evidence. Treat the calculations as predictions conditional on the stated pathway structure and assumptions.

Common pitfall: “The pulse ended, so the pathway output must immediately return to baseline.” This confuses input timing with downstream state. Each layer has its own time constant, so activity already in the cascade continues to evolve after the input switches off. Use the delayed step-off term $F_n(t-d)$, or solve the layer equations, to calculate the post-pulse response.

Check Your Understanding

Answer each question, then compare with the solutions.

A normalized one-layer pathway has $\tau=2$ minutes and receives a unit step input. What is its output after 2 minutes?
For a one-layer pathway, explain what happens to the periodic trough as the off interval becomes much longer than $\tau$.
A two-layer cascade has identical time constants. At the end of a pulse lasting exactly one time constant, is its normalized output greater or less than that of a one-layer cascade?
Why can two pulse trains with equal average input produce different peak and trough outputs?

Solutions. (1) The one-layer step response is $1-e^{-t/\tau}$, so at 2 minutes it is $1-e^{-1}=0.6321$ output units. (2) The trough is the peak multiplied by $e^{-L/\tau}$, where $L$ is the off interval. As $L/\tau$ becomes large, this factor approaches zero, so the trough approaches baseline. (3) It is less. At $t=\tau$, $F_1=1-e^{-1}=0.6321$, whereas $F_2=1-e^{-1}(1+1)=0.2642$. (4) The pathway retains activity between pulses, and peak and trough depend on pulse amplitude, duration, interval, and $\tau$, not solely on the time-averaged input.