Connecting Drug Exposure to Cascade Inputs and Outputs

From Drug Concentration to a Measured Endpoint

A drug concentration does not have to alter the measured biomarker immediately. It may first change the signal entering a cascade, after which successive stages transmit and delay that change. To interpret the observed response, it is useful to keep three elements distinct: the concentration-driven input, the dynamics of the cascade, and the measurement scale. A large endpoint change could reflect a strong input, efficient propagation through the stages, or a large measurement scale; those explanations are not interchangeable.

The model below treats concentration as the driver of an input to a linear, first-order cascade. It is a formulation for connecting exposure to the endpoint, not a claim that every drug acts through this exact mechanism. In particular, a constant-concentration step gives a tractable illustration of the connection. More complex exposure histories require the input to vary over time.

Definition: Let $C$ denote drug concentration and $u(C)$ the concentration-driven input to the first cascade stage, expressed as a deviation from the untreated input. For $m$ stages, let $x_j(t)$ be the deviation in stage $j$, with $x_0(t)=u(C(t))$. A first-order cascade is $$ \frac{dx_j(t)}{dt}=k_j\bigl(x_{j-1}(t)-x_j(t)\bigr), \qquad j=1,\ldots,m, $$ where each $k_j>0$ is a stage rate with units of inverse time. A linearly scaled measured endpoint is $Y(t)=Y_0+Sx_m(t)$, where $Y_0$ is the baseline endpoint and $S$ converts the final-stage deviation to endpoint units.

The input function specifies the magnitude and direction of the drug-driven signal. The rates $k_j$ specify how quickly each stage approaches its upstream value. The scale $S$ specifies how the final-stage signal is represented by the assay or endpoint. In this formulation, the stages have unit steady-state gain: if the input is held at a constant value $U$, each stage eventually approaches $U$. Thus stage rates govern the time course, not the final steady-state amplification.

Choosing a Concentration-Driven Input

A common saturable input is a signed Hill function:

u(C)=G\frac{C^h}{EC_{50}^{\,h}+C^h}.

Here $G$ is the maximum signed change in input, $EC_{50}$ is the concentration at which the fractional input reaches one half, and $h>0$ is the Hill coefficient. The concentration and $EC_{50}$ must use the same units, so the fraction is dimensionless. The input $u(C)$ therefore has the same units as the stage signals. A negative $G$ represents a decrease in the signal as concentration rises; a positive $G$ represents an increase. This sign convention separates direction from magnitude.

A linear input, $u(C)=\beta C$, is another possible formulation. It assumes that input changes in direct proportion to concentration over the range considered and does not impose a finite maximum. The Hill form instead approaches $G$ at high concentration. Choosing between them is a scientific and empirical decision: the function should reflect the intended exposure-response relationship over the observed concentration range, rather than being chosen solely because it fits one endpoint value.

Worked Example: Calculating a Saturable Input

Suppose a hypothetical drug produces an inhibitory input with $G=-5.00$ signal units, $EC_{50}=2.00$ mg/L, and $h=2$. At a constant concentration of $C=4.00$ mg/L, calculate the input.

Substitute into the Hill function:

u(4.00)=-5.00\frac{(4.00)^2}{(2.00)^2+(4.00)^2} =-5.00\frac{16.0}{4.00+16.0} =-5.00(0.800) =-4.00\text{ signal units}.

The hypothetical concentration therefore produces an input decrease of 4.00 signal units relative to the untreated input. The result is smaller in magnitude than the maximum decrease of 5.00 units, as expected for a finite concentration under this saturable function. It is an input value; it is not yet the predicted endpoint change, because the cascade and measurement scale still need to be applied.

How Stage Rates Shape the Endpoint Time Course

For a constant concentration switched on at time zero, let the resulting input be a constant $U$, with all stage deviations initially zero. The first stage approaches $U$ exponentially. The downstream stages then respond to their upstream stages, so their trajectories lag. For two stages with unequal rates, the equations are

\frac{dx_1}{dt}=k_1(U-x_1), \qquad \frac{dx_2}{dt}=k_2(x_1-x_2), \qquad x_1(0)=x_2(0)=0.

Solving the first equation gives $x_1(t)=U(1-e^{-k_1t})$. Substituting this into the second equation and multiplying by the integrating factor $e^{k_2t}$ gives

\frac{d}{dt}\bigl(e^{k_2t}x_2(t)\bigr) =k_2U\bigl(e^{k_2t}-e^{(k_2-k_1)t}\bigr).

Integrating from $0$ to $t$, and using $x_2(0)=0$, yields the unequal-rate step response:

x_2(t) =U\left[1-\frac{k_2e^{-k_1t}-k_1e^{-k_2t}}{k_2-k_1}\right], \qquad k_1\ne k_2.

The expression tends to zero at $t=0$ and to $U$ as $t$ increases without bound. Both rates affect how quickly the endpoint stage approaches its limiting value, but neither changes that limiting value in this unit-gain cascade. If the rates are equal, the unequal-rate expression is not used by direct substitution because its denominator becomes zero; the equal-rate limit is $x_2(t)=U[1-(1+kt)e^{-kt}]$, where $k_1=k_2=k$.

Worked Example: Two Unequal Transit Rates

Suppose a hypothetical constant input of $U=2.00$ signal units enters a two-stage cascade at time zero. Let $k_1=1.00$ h$^{-1}$, $k_2=2.00$ h$^{-1}$, and calculate the second-stage deviation after $t=1.00$ h. Assume both stages begin at zero deviation and obey the stated first-order equations.

Because the rates are unequal, apply the derived response:

x_2(1.00) =2.00\left[1-\frac{2.00e^{-1.00(1.00)}-1.00e^{-2.00(1.00)}}{2.00-1.00}\right] =2.00\left(1-2e^{-1}+e^{-2}\right).

Using $e^{-1}=0.3678794$ and $e^{-2}=0.1353353$, the fraction of the eventual stage response is

1-2(0.3678794)+0.1353353 =0.3995765 \approx 0.3996.

Therefore,

x_2(1.00)=2.00(0.3995765)=0.7991530 \approx 0.7992\text{ signal units}.

After one hour, the second stage has reached about 39.96% of its eventual change, which is $2.00$ signal units. The result is less than the input because the cascade has not reached steady state. The different rates influence the transient response; the eventual second-stage deviation remains $2.00$ signal units.

From Final-Stage Signal to Measured Endpoint

The measurement equation $Y(t)=Y_0+Sx_m(t)$ makes the scale and baseline explicit. $Y_0$ has endpoint units, $x_m$ has signal units, and $S$ has endpoint units per signal unit. Consequently, a positive final-stage deviation increases the measured endpoint when $S>0$; if the assay orientation or measurement definition makes $S<0$, the endpoint changes in the opposite direction. The units and sign should be checked before interpreting a fitted coefficient.

Worked Example: Applying the Measurement Scale

Use the second-stage result from the preceding hypothetical calculation. Suppose the measured endpoint has baseline $Y_0=10.0$ endpoint units and scale $S=3.00$ endpoint units per signal unit. Calculate the endpoint at one hour.

First calculate the endpoint change:

\Delta Y(1.00)=Sx_2(1.00) =(3.00\text{ endpoint units/signal unit})(0.7991530\text{ signal units}) =2.397459\text{ endpoint units} \approx 2.397\text{ endpoint units}.

Then add the baseline:

Y(1.00)=Y_0+\Delta Y(1.00) =10.0+2.397459 =12.397459 \approx 12.40\text{ endpoint units}.

The predicted endpoint is about 12.40 units, an increase of about 2.397 units from baseline. Reporting the endpoint change separately from the endpoint value helps prevent confusion between the modeled signal deviation and the measured quantity.

This decomposition also reveals a scaling limitation. At a fixed concentration, if the endpoint depends on input gain only through the product $SG$, endpoint observations alone may not distinguish a larger input gain from a larger measurement scale. For example, $S=3$ endpoint units per signal unit and $G=2$ signal units give the same product $SG=6$ as $S=2$ endpoint units per signal unit and $G=3$ signal units. Without independent calibration or additional information, both parameter pairs can imply the same endpoint trajectory under the same stage dynamics. The numerical values of $S$ and $G$ should not then be interpreted as separately identified merely because a model-fitting procedure returns estimates for both.

Assumptions, Diagnostics, and Alternatives

The response calculations rely on several substantive assumptions. Concentration is treated as the driver of $u(C)$; the chosen concentration-input function is correct over the range used; each stage follows a linear first-order equation with a positive rate; transfer between stages has unit gain; and the measured endpoint is a linear scaling of the final-stage deviation plus a stable baseline. For a step-response calculation, the concentration-induced input is constant after the switch, and the initial stage deviations are specified. If any of these assumptions fail, the calculated trajectory may not describe the observed endpoint even when the arithmetic is correct.

A useful diagnostic is to compare predicted and observed response direction, timing, and limiting level. Under a constant positive input step and the stated positive-rate cascade, each stage moves toward the same input level; the model does not generate a sustained overshoot beyond that level. A reproducible overshoot, an endpoint that moves in the wrong direction, or a steady-state change inconsistent with $S U$ suggests checking the input sign, baseline, endpoint scaling, initial conditions, and model structure. Such patterns do not by themselves identify which assumption is wrong.

Other warning signs include concentration values outside the range that can inform the chosen input function, estimates of an input or measurement scale that depend strongly on arbitrary units, and residual patterns that track time or concentration. For example, a linear input may predict increasingly large changes at high concentration even when the observed response appears to plateau. Conversely, a saturable input can be poorly distinguished from a linear one when all observed concentrations are far below $EC_{50}$. These are limitations of information in the data, not problems fixed by reporting more decimal places.

The main alternatives make different assumptions. A direct concentration-to-endpoint model treats the measured endpoint as responding without the modeled cascade delay. A linear concentration input assumes proportionality over the relevant range, while the Hill input assumes saturation. The first-order cascade adds stage-mediated delay while retaining linear transmission between stages. Each can be useful under its own assumptions; they should be compared using the response pattern and scientific rationale, not by treating a more elaborate cascade as automatically more realistic. The present formulation also does not account for nonlinear transfer between stages, time-varying stage properties, or an endpoint that is a nonlinear function of the final-stage signal.

Common mistake: Treating the measured endpoint change as if it were the concentration-driven input. The input is $u(C)$; the endpoint change is $Sx_m(t)$. In general, $x_m(t)$ depends on the stage rates, the number and arrangement of modeled stages, the input history, and the initial conditions. The correction is to calculate the input first, propagate it through the stated cascade, and apply the measurement scale only to the final-stage signal.

The practical value of this separation is interpretability. It allows an analyst to ask whether a mismatch comes from the concentration-response relation, the speed of signal propagation, or the mapping from signal to assay units. It also makes units a check on model construction: stage rates must have inverse-time units, the Hill fraction must be dimensionless, and $Sx_m$ must have endpoint units. Keeping those roles distinct is essential before interpreting any fitted parameter as a biological property.

Check Your Understanding

Try each question, then compare your reasoning with the solutions.

With $G=4.00$ signal units, $EC_{50}=3.00$ mg/L, $h=1$, and $C=3.00$ mg/L, what is $u(C)$?
In a unit-gain cascade driven by a constant input $U$, what is the limiting value of the final stage, and which parameters govern its transient approach?
If $x_m=0.500$ signal units, $Y_0=8.00$ endpoint units, and $S=-2.00$ endpoint units per signal unit, calculate $Y$ and interpret the direction of change.
Why can endpoint data at one concentration fail to distinguish input gain $G$ from measurement scale $S$?

Solutions. (1) The Hill fraction is $3.00/(3.00+3.00)=0.500$, so $u(C)=4.00(0.500)=2.00$ signal units. (2) The final stage tends to $U$, because each stage has unit steady-state gain; the rates $k_j$ govern the transient approach. (3) $Y=8.00+(-2.00)(0.500)=7.00$ endpoint units, a decrease of 1.00 endpoint unit from baseline. (4) If the endpoint response depends on the product $SG$, different values of $S$ and $G$ with the same product can produce the same endpoint response, so independent calibration or other informative data are needed to separate them.