Why Exposure and Response May Not Peak Together
In an early-phase study, plasma concentration may rise and fall faster than a pharmacodynamic biomarker can respond. A plot of biomarker response against plasma concentration can then trace a loop rather than a single curve: the biomarker differs on the rising and falling concentration limbs, even at the same plasma concentration. This pattern is called hysteresis. It is a feature of the observed time course, not by itself proof that plasma concentration causes the response or that a particular delay mechanism is operating.
An effect compartment is a useful empirical model for one common pattern: plasma concentration changes first, and the concentration driving the response approaches it gradually. It introduces a latent effect-site concentration, \(C_e\), which is not necessarily a directly measured tissue concentration. The model can represent a delay between plasma exposure and response without specifying the detailed biological processes that create that delay.
Direction of the Loop
Read the loop in chronological order, with plasma concentration on the horizontal axis and response on the vertical axis. In a counterclockwise loop, the response at a given concentration is generally lower on the rising limb than on the falling limb. This is the pattern expected when the response lags behind plasma concentration: as plasma concentration rises, the effect compartment has not yet caught up; after plasma concentration falls, the effect compartment can remain elevated. A simple effect-compartment model therefore commonly produces counterclockwise hysteresis.
In a clockwise loop, response is generally higher on the rising limb than on the falling limb at the same plasma concentration. This can occur when the response appears to lead plasma concentration, or when responsiveness changes over time, for example through acute tolerance or another time-varying process. The direction is descriptive. A clockwise loop is not explained by a simple effect compartment that only lags plasma concentration; fitting that model does not make the mismatch disappear.
A loop may be difficult to classify when observations are sparse, noisy, or do not cover both rising and falling exposure. The comparison at a given concentration should be based on observations from the same participant and dosing occasion when possible. Pooling observations across people with different baselines, exposure histories, or response sensitivities can create an apparent loop that does not represent an individual time course.
Worked Example: Classifying a Counterclockwise Loop
Suppose a hypothetical participant has the following time-ordered observations. The response is a biomarker change in units; plasma concentration is in mg/L.
| Time (h) | Plasma concentration (mg/L) | Biomarker response (units) | Phase |
|---|---|---|---|
| 1 | 5 | 2 | Rising |
| 2 | 10 | 4 | Rising |
| 4 | 10 | 7 | Falling |
| 6 | 5 | 5 | Falling |
At 10 mg/L, response is 4 units on the rising limb and 7 units on the falling limb. At 5 mg/L, it is 2 units on the rising limb and 5 units on the falling limb. Thus, at both matched concentrations, the falling-limb response is higher. Traversing the points in time order gives the usual counterclockwise pattern. These four observations suggest hysteresis, but do not establish its mechanism or estimate a reliable delay by themselves.
The Effect-Compartment Model
The model assumes that the rate of change in \(C_e\) is proportional to the difference between plasma and effect-compartment concentrations. The proportionality constant \(k_{e0}\) is positive and has units of inverse time. For clarity, the basic form assumes that plasma concentration is the input driving equilibration, that \(k_{e0}\) is constant over the modeled period, and that one effect-compartment concentration is sufficient to represent the delay.
When \(C_p\) is held constant at \(C\) over an interval of length \(\Delta t\), this equation has a simple exact solution. Subtract \(C\) from both sides by defining \(D(t)=C_e(t)-C\). Then \(dD/dt=-k_{e0}D\), so \(D(t+\Delta t)=D(t)\exp(-k_{e0}\Delta t)\). Substituting back gives the update used for piecewise-constant plasma input:
This update moves \(C_e\) toward \(C\) without overshooting it during the interval. A larger \(k_{e0}\) means faster equilibration and less lag; a smaller \(k_{e0}\) means slower equilibration and a longer half-time. If plasma concentration changes continuously within an interval, it is not generally exact to treat the entire interval as having one constant concentration. The input profile must instead be represented at an adequate time resolution or integrated using the specified concentration-time function.
To connect \(C_e\) with response, a concentration-response function can be applied at each time. For a hypothetical increasing biomarker response, one possible model is the Emax function:
Here \(E_0\) is the response at zero effect-compartment concentration, \(E_{\max}\) is the maximum increase above \(E_0\), and \(EC_{50}\) is the value of \(C_e\) producing half of that maximum increase. This response function is a modeling choice, not a necessary part of the effect-compartment definition. For an inhibitory response, the direction and form of the concentration-response function must be appropriate to the biomarker.
Worked Example: Updating the Effect-Compartment Concentration
Consider hypothetical concentrations held constant over four successive one-hour intervals: 10, 20, 10, and 0 mg/L. Let \(C_e(0)=0\) mg/L and \(k_{e0}=0.50\ \mathrm{h}^{-1}\). The update factor is \(\exp(-0.50\times1)=0.6065\). Applying the recurrence consistently gives:
The recurrence shows the lag: after the interval at 20 mg/L, \(C_e\) remains about 10.26 mg/L during the next interval even though plasma concentration is 10 mg/L, and it remains about 6.159 mg/L after the input falls to zero. Carrying the updates at full precision gives \(C_{e,1}=3.93469\), \(C_{e,2}=10.25590\), \(C_{e,3}=10.15521\), and \(C_{e,4}=6.15945\); to three decimals, these are 3.935, 10.256, 10.155, and 6.159 mg/L. The equilibration half-time is \(\ln(2)/0.50=1.386\) h.
For an illustrative response model with \(E_0=2\) units, \(E_{\max}=8\) units, and \(EC_{50}=5\) mg/L, the first and third effect-compartment concentrations correspond to responses of approximately \(2+8(3.935)/(5+3.935)=5.523\) units and \(2+8(10.155)/(5+10.155)=7.361\) units. The plasma input is 10 mg/L in both intervals, but the modeled response differs because the effect-compartment concentration depends on the preceding exposure history.
Estimating the Delay from a Profile
In practice, \(k_{e0}\) is estimated by fitting the combined exposure-response model to the time-ordered data. For each candidate value of \(k_{e0}\), the effect-compartment trajectory is calculated from the plasma profile and an initial condition, then converted to predicted biomarker responses. The candidate value is judged by how well those predictions match the observed responses, while other model parameters may also need to be estimated. A single best-fitting value is not automatically precise or biologically unique.
Worked Example: Comparing Candidate Equilibration Rates
Use the hypothetical one-hour plasma inputs 10, 20, 10, and 0 mg/L, with \(C_e(0)=0\), and the response function \(R=2+8C_e/(5+C_e)\). Suppose observed responses at the ends of the four intervals are 5.6, 7.2, 7.3, and 6.3 units. To illustrate estimation, compare two candidate rates using the sum of squared residuals, \(SSE=\sum_i(R_i-\widehat R_i)^2\). Assume for this simple comparison that \(E_0\), \(E_{\max}\), and \(EC_{50}\) are fixed at the stated values.
For \(k_{e0}=0.50\ \mathrm{h}^{-1}\), the recurrence gives \(C_e\) values of approximately 3.935, 10.256, 10.155, and 6.159 mg/L. The corresponding predictions are approximately 5.523, 7.378, 7.361, and 6.416 units. Using those displayed predictions:
For \(k_{e0}=1.00\ \mathrm{h}^{-1}\), the update factor is \(\exp(-1)=0.3679\). The resulting \(C_e\) values are approximately 6.321, 14.97, 11.83, and 4.351 mg/L, giving predictions of approximately 6.467, 7.997, 7.623, and 5.722 units. The squared-error calculation is:
Among these two candidates, \(0.50\ \mathrm{h}^{-1}\) fits better because its SSE is smaller. Its corresponding half-time is 1.386 h. This is only a coarse grid comparison, not a complete parameter-estimation analysis: a formal fit would search across a range of \(k_{e0}\), account for the observation-error model, and usually consider uncertainty and the joint estimation of response parameters. The small example demonstrates the calculation, not the strength of evidence available from four observations.
Assumptions, Diagnostics, and Alternatives
The effect-compartment interpretation depends on the exposure input, timing, and response model being adequate. Important checks include whether the concentration-time profile is sampled densely enough to capture relevant changes, whether response measurements are correctly aligned in time, and whether baseline drift or time-varying controls could explain the apparent loop. The model also assumes a stable \(k_{e0}\) over the modeled period and a consistent relationship between \(C_e\) and response. These are assumptions to assess, not conclusions guaranteed by fitting.
Signs of a poor fit include systematic residual patterns over time, predicted peaks that occur too early or too late, a fitted rate near a search boundary, or markedly different rates when a small number of observations are removed. A loop that changes direction, or a response that is not a single-valued function of \(C_e\), may indicate that one lag compartment is insufficient. When the data contain little information on both the rising and falling limbs, many rates can produce similar predictions; the delay may then be weakly identified.
An effect-compartment model is not interchangeable with every model of delayed response. A turnover model, covered in the earlier material on turnover models, represents response change through production and loss of the biomarker. A direct-effect model relates response to contemporaneous concentration and can be a reasonable alternative when there is little evidence of a time lag. The effect-compartment approach instead transforms the exposure history into a delayed concentration driver and then applies a concentration-response function. Comparing alternatives should be guided by the observed time course, biological plausibility, model fit, and whether the data can distinguish the structures.
The practical value of the model is that it can prevent a delayed biomarker from being misread as an inconsistent or absent exposure-response relationship. Its limitation is equally important: a good fit does not identify the biological source of the delay, and sparse early-phase data may not support a precise delay estimate. State the observed loop direction, the model structure and assumptions, the estimated rate and half-time with units, and the limitations that affect interpretation.
Check Your Understanding
Answer each question before reviewing the brief solutions.
- At the same plasma concentration, response is lower on the rising limb than on the falling limb. What loop direction is expected, and what does this pattern suggest?
- For \(k_{e0}=0.25\ \mathrm{h}^{-1}\), calculate the equilibration half-time.
- Starting from \(C_e=0\) mg/L, plasma concentration is held at 8 mg/L for one hour and \(k_{e0}=0.50\ \mathrm{h}^{-1}\). What is the effect-compartment concentration after that hour?
- Why does a counterclockwise loop not by itself prove that a tissue concentration delay caused the response pattern?
Solutions. (1) Counterclockwise; it is consistent with response lagging plasma concentration, but does not establish the mechanism. (2) \(t_{1/2,e0}=\ln(2)/0.25=2.773\) h. (3) \(C_e=8+(0-8)\exp(-0.50)=8(1-0.6065)=3.148\) mg/L. (4) Other time-dependent processes, including turnover, changing sensitivity, baseline drift, and sampling variation, can produce or contribute to the loop.