Why Exposure and Response May Not Move Together
A small-molecule concentration can rise and fall more quickly than the pathway response it produces. Consequently, the same measured exposure may be associated with different responses at different times: one response while exposure is increasing and another while it is decreasing. This time dependence is called hysteresis. Its direction is informative, but the loop alone does not identify which biological process caused the delay.
To interpret direction consistently, this tutorial uses exposure on the horizontal axis and an increasing measure of pharmacologic effect on the vertical axis. For example, the response may be fractional inhibition, or the magnitude of a biomarker change from baseline. With this convention, a delayed positive effect is generally greater on the declining-exposure branch than on the rising-exposure branch, producing a counterclockwise loop. If the vertical axis instead shows the raw value of a biomarker that decreases with treatment, the apparent direction can reverse.
The relevant exposure might be unbound concentration in plasma, \(C_u(t)\), or unbound concentration at the effect site, \(C_e(t)\). Comparing response with both, when the measurements or model estimates are available, helps locate where a delay enters. A loop against plasma exposure may reflect distribution to the effect site, target binding, downstream biomarker turnover, or a combination. A loop against effect-site concentration rules out neither binding nor downstream delay, but it removes effect-site equilibration as the direct explanation for that comparison.
Three Sources of Delay
The following simple models clarify the distinctions. They are diagnostic representations, not claims that every compound or pathway follows these equations exactly. They assume that the measured time courses are aligned, the relevant exposure is represented adequately, and major determinants such as dose, substrate, and baseline physiology do not change in unmodeled ways during the comparison.
First, effect-site equilibration can be represented as a first-order approach of \(C_e\) to \(C_u\):
Here \(k_{e0}\) has units of inverse time, and \(C_e\) and \(C_u\) have concentration units. If the target responds instantaneously to \(C_e\), a loop against \(C_u\) may occur even though response is a single-valued function of \(C_e\).
Second, target binding can itself be dynamic. For a simple reversible one-site interaction, let \(\theta\) be the fraction of target occupied, \(k_{\mathrm{on}}\) the association rate constant, and \(k_{\mathrm{off}}\) the dissociation rate constant:
The units of \(k_{\mathrm{on}}\) are concentration\(^{-1}\) time\(^{-1}\), while \(k_{\mathrm{off}}\) has units of inverse time. At constant \(C_e\), the steady-state occupancy is \(C_e/(C_e+K_D)\), where \(K_D=k_{\mathrm{off}}/k_{\mathrm{on}}\). If binding and unbinding take time, occupancy can lag even when \(C_e\) is known.
Third, a downstream biomarker may turn over more slowly than target activity changes. One illustrative production-loss model is:
Here \(B\) is the biomarker amount or concentration, \(f(\theta)\) is a dimensionless production multiplier, and \(k_{\mathrm{out}}\) has units of inverse time. Since \(f(\theta)\) is dimensionless, \(k_{\mathrm{in}}\) has units of biomarker per time. For example, \(f(\theta)=1-I_{\max}\theta\) describes inhibition of production when \(0\leq I_{\max}\theta\leq1\). At baseline, with no drug and \(f=1\), this model has steady-state biomarker \(B_0=k_{\mathrm{in}}/k_{\mathrm{out}}\).
Worked Examples: Reading Direction and Locating Delay
Worked Example: Determining Loop Direction From a Time Course
Consider hypothetical measurements of exposure in nM and fractional inhibition, with time increasing down the table. Assume exposure increases and then decreases, and use fractional inhibition as the vertical-axis response.
| Time (h) | Exposure (nM) | Fractional inhibition |
|---|---|---|
| 0 | 0 | 0.00 |
| 1 | 4 | 0.10 |
| 2 | 8 | 0.35 |
| 3 | 5 | 0.50 |
| 4 | 1 | 0.20 |
| 5 | 0 | 0.05 |
Exposure of 5 nM occurs on the rising branch between 4 and 8 nM. Linear interpolation between the time-1 and time-2 observations gives the rising-branch inhibition:
On the falling branch, the observed inhibition at 5 nM is 0.50. Thus the response is higher at the same exposure during decline than during rise. With exposure on the horizontal axis and inhibition increasing upward, the loop is counterclockwise. The inhibition also peaks after the exposure peak, consistent with a delay, but these observations alone do not identify its source.
Worked Example: Testing for Effect-Site Distribution Delay
Suppose a hypothetical unbound plasma concentration steps from 0 to 10 nM and remains there for 2 hours. Let \(k_{e0}=0.5\ \mathrm{h}^{-1}\) and let the initial effect-site concentration be 0. For a constant step input, solving the effect-site equation gives \(C_e(t)=C_u(1-e^{-k_{e0}t})\). At 2 hours:
If plasma exposure then returns to zero, the effect-site concentration declines as \(C_e(t)=C_e(2)e^{-k_{e0}(t-2)}\). Two hours after the decline begins, \(C_e(4)=6.321e^{-1}\approx2.325\ \mathrm{nM}\), while \(C_u(4)=0\). This persistence at the effect site can generate a loop against plasma exposure. Under an instantaneous effect-site response with \(E=C_e/(5+C_e)\), the effect at 2 hours is \(6.321/(5+6.321)\approx0.558\), or 55.8% of the maximum, and it remains nonzero after plasma concentration has returned to zero. A response loop that substantially collapses when plotted against \(C_e\), rather than \(C_u\), supports distribution delay as an explanation.
Worked Example: Checking for Slow Target Binding
Suppose hypothetical effect-site concentration is held at 10 nM, with \(k_{\mathrm{on}}=0.10\ \mathrm{nM}^{-1}\mathrm{h}^{-1}\), \(k_{\mathrm{off}}=0.50\ \mathrm{h}^{-1}\), and initial occupancy \(\theta(0)=0\). The equilibrium dissociation constant is:
The steady-state occupancy at 10 nM is \(10/(10+5)=0.6667\). For constant concentration, the binding equation has rate constant \(k_{\mathrm{on}}C_e+k_{\mathrm{off}}=1.50\ \mathrm{h}^{-1}\), so:
At 1 hour, \(\theta(1)=0.6667(1-e^{-1.50})\approx0.5179\), or 51.79%. At 2 hours, \(\theta(2)=0.6667(1-e^{-3.00})\approx0.6335\), or 63.35%. Both are below the 66.67% equilibrium occupancy. Thus, even if effect-site concentration has equilibrated, occupancy can still be increasing. A loop against \(C_e\) together with a lagging occupancy or target-activity time course is consistent with slow target binding.
Worked Example: Identifying Downstream Biomarker Turnover
Suppose a hypothetical biomarker has baseline \(B_0=100\) units and \(k_{\mathrm{out}}=0.20\ \mathrm{h}^{-1}\). Then \(k_{\mathrm{in}}=k_{\mathrm{out}}B_0=20\) units per hour. Assume target inhibition immediately and persistently reduces the production multiplier to \(f(\theta)=0.40\), with the biomarker initially at baseline. The treated steady state is:
For a constant production multiplier, \(B(t)=B_{\mathrm{ss}}+(B(0)-B_{\mathrm{ss}})e^{-k_{\mathrm{out}}t}\). At 3 hours:
The target-level effect has already changed, but the biomarker remains at about 72.93 units rather than its eventual 40 units. A biomarker loop that persists when plotted against measured target activity, with a delay matching biomarker turnover, supports a downstream turnover explanation.
A Practical Diagnostic Sequence
A useful diagnosis compares successive links in the time course rather than assigning a mechanism from loop direction alone. This is most informative when exposure, target engagement or activity, and biomarker are sampled densely enough to establish their relative timing. If one of these quantities is not measured, a fitted mechanistic model can be considered, but its conclusions depend on the model assumptions and available data.
State whether exposure is \(C_u\) or \(C_e\), whether response means inhibition, activation, or raw biomarker value, and which branch is rising or falling. Check that the apparent loop is not an artifact of reversing response coding.
A loop against \(C_u\) that diminishes against \(C_e\) is consistent with effect-site equilibration. It does not by itself prove that distribution is the only process involved.
If target engagement still lags \(C_e\), dynamic binding or another target-level process may be needed. Check whether the estimated binding rates are supported by the time-course data.
If target activity changes before the biomarker, downstream turnover is plausible. A turnover model should explain both the biomarker's direction of change and its recovery after target activity returns.
Assess sampling times, assay delays, baseline drift, active metabolites, changing substrate, and treatment-related feedback. If several delays could explain the observations, report that ambiguity rather than selecting one mechanism from the loop alone.
Direction, Alternatives, and Limitations
With the convention used here, a delayed positive response tends to be lower at a given exposure during the rising phase and higher during the falling phase. That pattern is counterclockwise. A clockwise loop can indicate that response leads exposure, as may occur when an effect changes quickly but concentration decays slowly, or it may arise from other time-varying processes. If the vertical axis is a raw biomarker that falls as inhibition increases, the same biological delay can look clockwise instead. Therefore, “clockwise means tolerance” or “counterclockwise proves distribution delay” is not a reliable general rule.
The main alternatives answer different questions. An instantaneous exposure-response relationship assumes response is determined by the selected exposure without appreciable lag. An effect-site model adds a distribution compartment and is useful when response tracks \(C_e\) more closely than plasma concentration. A dynamic binding model represents target occupancy changing over time at a given \(C_e\). A turnover model represents delayed change in the biomarker after its production or loss rate changes. These mechanisms can coexist; fitting only one when several are active can produce misleading parameter estimates.
Common diagnostics include plotting time-ordered observations with arrows, checking whether the same exposure was sampled on both rising and falling phases, and plotting response against each candidate upstream variable. A loop may be poorly resolved if sampling is sparse, if there is little exposure decline before the last observation, or if measurement error is large relative to the branch separation. Hysteresis can also reflect a time-varying baseline, active metabolites, feedback, tolerance, or assay processing delay. A loop is evidence that the chosen exposure alone may not describe response over time; it is not proof of a specific causal mechanism.
Check Your Understanding
Answer each question using the stated axis conventions. Brief solutions follow the questions.
- With exposure on the horizontal axis and fractional inhibition increasing upward, inhibition is lower on the rising branch than on the falling branch at the same exposure. What is the loop direction?
- A response loop against plasma exposure nearly disappears when response is plotted against effect-site concentration. Which delay is supported, and what does this observation not establish?
- At constant effect-site concentration, why can occupancy remain below its equilibrium value immediately after exposure increases?
- In a production-loss biomarker model, what are the units of \(k_{\mathrm{in}}\) when \(f(\theta)\) is dimensionless and \(B\) is measured in concentration units?
- Give one reason a hysteresis loop might not identify distribution, binding, or turnover as the unique cause.
Solutions. (1) Counterclockwise under the stated axes: the path moves right along the lower branch as exposure rises and returns left along the upper branch as exposure falls. (2) Effect-site equilibration is supported; the loop does not prove it is the only delay or exclude other processes. (3) Binding and unbinding take finite time, so occupancy approaches rather than instantaneously reaches its steady state. (4) Concentration per unit time, because the production term \(k_{\mathrm{in}}f(\theta)\) must have those units. (5) Several mechanisms can coexist or produce similar time courses; sparse sampling, feedback, active metabolites, or baseline drift can also confound interpretation.