Dose Fractionation and Payload Release Timing in ADCs

Why Dose Timing Matters

Two ADC regimens can deliver the same total dose but produce different intracellular payload profiles. Splitting a dose may lower the maximum concentration of active payload, yet delay that maximum because each dose must pass through internalization and release steps. Whether this schedule change also alters cumulative exposure depends on the model: in a linear system with complete payload recovery and clearance, total AUC depends on total dose, not on how that dose is divided.

The distinction matters when peak concentration and cumulative exposure have different consequences. A high peak may be relevant to concentration-dependent toxicity, while cumulative exposure may better represent effects driven by time-integrated payload. Neither measure alone determines clinical outcome; translating exposure into efficacy or toxicity requires additional assumptions and data.

A Delayed-Release Model

Consider three tumor-associated states following a dose: $S$, ADC available for internalization; $I$, internalized ADC carrying unreleased payload; and $C$, active intracellular payload concentration. The model treats internalization, payload release, and active-payload loss as first-order processes. For clarity, use rates $k_i=0.25\ \mathrm{h}^{-1}$, $k_r=0.50\ \mathrm{h}^{-1}$, and $k_c=0.50\ \mathrm{h}^{-1}$. A dose $D_j$ at time $t_j$ enters $S$; between doses, the states obey:

\frac{dS}{dt}=-k_iS,\qquad \frac{dI}{dt}=k_iS-k_rI,\qquad \frac{dC}{dt}=k_rI-k_cC.

These equations describe a deliberately compact model: internalization moves ADC from $S$ to $I$, release moves payload into $C$, and active payload is lost at rate $k_c$. The dose-to-state scaling is absorbed into the units of the input and response. Thus, the response kernel below is expressed in intracellular concentration per dose unit. There is no direct instantaneous jump in $C$ at dosing.

Definition: The dose-response kernel $h(u)$ is the intracellular active-payload concentration $u$ hours after one unit of dose, with $h(u)=0$ for $u<0$. In this model, $u$ is time since the dose and the rates are constant. For doses $D_j$ given at times $t_j$, the predicted concentration is the sum of their separate responses: $$ C(t)=\sum_j D_j h(t-t_j),\qquad t\geq t_j\ \text{for the contributing doses}. $$

For the stated rates, solving the linear equations for a single dose gives:

h(u)=2e^{-0.25u}-(2+0.5u)e^{-0.5u},\qquad u\geq 0.

Although this expression contains a subtraction, it is nonnegative for $u\geq0$: it is the concentration response of the serial compartment model, not a difference between two independently dosed payloads. It starts at zero, rises as internalized ADC releases payload, and eventually declines as the slower upstream pool empties and active payload is cleared. The values of $h$ have units of concentration per dose unit; $u$ is in hours.

Peak Concentration and Cumulative Exposure

For one dose, the peak occurs where the rate of change of the kernel is zero. Differentiating gives:

h'(u)=-0.5e^{-0.25u}+(0.5+0.25u)e^{-0.5u}.

The peak time therefore solves $(1+0.5u)e^{-0.25u}=1$. The kernel increases before this root and decreases after it, so the root identifies its maximum. For multiple doses, the peak must be found from the derivative of the sum of the shifted responses; it need not occur at the single-dose peak time after any one dose.

Cumulative exposure over a finite interval also follows directly from the kernel. For a single dose, define $F(T)=\int_0^T h(u)\,du$. Integrating each exponential term gives:

F(T)=2-8e^{-0.25T}+(6+T)e^{-0.5T}, \qquad F(\infty)=\int_0^\infty h(u)\,du=2.

Consequently, if every dose has the same disposition and the observation interval includes the complete response to each dose, total AUC is $2\sum_jD_j$, regardless of dose timing. A finite observation window is different: doses given later contribute less of their eventual AUC before the window ends.

Worked Example: Peak After One Dose

Suppose a hypothetical ADC regimen gives a single dose of $100\ \mathrm{mg/m^2}$. Use the stated kernel and rates to estimate the time and concentration of the intracellular payload peak. Assume the model is applicable over the concentration range reached and that the dose-to-tumor input is represented by the kernel.

The peak time solves $(1+0.5u)e^{-0.25u}=1$. Numerically, the positive solution is approximately $u=5.03\ \mathrm{h}$. Substituting into the kernel gives $h(5.03)\approx0.2036\ \mathrm{nM}/(\mathrm{mg/m^2})$. Thus:

C_{\max}=100(0.2036)\approx20.36\ \mathrm{nM}.

The model predicts a peak of about $20.4\ \mathrm{nM}$, roughly five hours after dosing. The delay follows from the serial internalization and release steps; it is not evidence that the administered ADC itself peaks at that time.

Worked Example: Two Fractionated Doses

Now divide the same total dose into two $50\ \mathrm{mg/m^2}$ doses at $t=0$ and $t=10\ \mathrm{h}$. Assume linear kinetics, identical access to the tumor for both doses, and no change in antigen availability or payload clearance during the interval. Superposition gives $C(t)=50h(t)+50h(t-10)$ for $t\geq10\ \mathrm{h}$.

The combined peak is obtained by setting $50h'(t)+50h'(t-10)=0$. Solving this equation numerically gives a peak near $t=14.41\ \mathrm{h}$. To show the arithmetic, evaluate the two contributions at $t=14.4\ \mathrm{h}$, essentially the same time at the precision used here. The first dose is $14.4$ hours old:

50h(14.4) =50\left[2e^{-0.25(14.4)}-(2+0.5(14.4))e^{-0.5(14.4)}\right] \approx50(0.0477789)=2.389\ \mathrm{nM}.

The second dose is $4.4$ hours old:

50h(4.4) =50\left[2e^{-0.25(4.4)}-(2+0.5(4.4))e^{-0.5(4.4)}\right] \approx50(0.200369)=10.018\ \mathrm{nM}.

The sum is $2.389+10.018=12.407\ \mathrm{nM}$, or about $12.41\ \mathrm{nM}$. The single $100\ \mathrm{mg/m^2}$ dose peaked at about $20.36\ \mathrm{nM}$, so this fractionated schedule lowers the predicted peak by about $7.95\ \mathrm{nM}$, or $39.0\%$ relative to the single-dose peak. Its peak occurs after the second dose because the remaining payload response from the first dose overlaps with the delayed response to the second.

For a complete observation of both responses, the AUC is unchanged:

\mathrm{AUC}_{0,\infty} =50F(\infty)+50F(\infty) =50(2)+50(2)=200\ \mathrm{nM\cdot h}.

This is the same as $100F(\infty)=200\ \mathrm{nM\cdot h}$ for the single dose. The model predicts a different peak and peak time, but not a different total AUC.

Worked Example: A Finite AUC Window

Compare cumulative exposure during the first $24\ \mathrm{h}$ after the first dose for the single $100\ \mathrm{mg/m^2}$ dose and the fractionated $50+50\ \mathrm{mg/m^2}$ schedule above. Use the same linear model. The finite-window integral for one dose is $F(T)$ as defined earlier.

For the single dose, $F(24)=2-8e^{-6}+30e^{-12}\approx1.98035\ \mathrm{h}$ in kernel-scaled units. Therefore:

\mathrm{AUC}_{0,24}^{\mathrm{single}} =100F(24) \approx100(1.98035) =198.0\ \mathrm{nM\cdot h}.

For the fractionated regimen, the first dose contributes through age $24\ \mathrm{h}$, while the second contributes through age $14\ \mathrm{h}$. Since $F(14)=2-8e^{-3.5}+20e^{-7}\approx1.77666$:

\mathrm{AUC}_{0,24}^{\mathrm{fractionated}} =50F(24)+50F(14) =50(1.98035+1.77666) \approx187.9\ \mathrm{nM\cdot h}.

The fractionated schedule has about $10.2\ \mathrm{nM\cdot h}$ less exposure inside this particular $24$-hour window, even though its eventual total AUC is also $200\ \mathrm{nM\cdot h}$. The difference is not lost exposure in the model; more of the second dose's response occurs after hour 24.

Assumptions, Diagnostics, and Alternatives

The schedule comparison is valid only under assumptions that make superposition appropriate. The equations use constant first-order rates; dose responses add without competition; the tumor input per unit dose is the same at each administration; and dosing does not change target abundance, internalization, release, or payload clearance. The model also assumes that the same measured intracellular concentration represents the relevant payload across schedules. These are modeling assumptions, not guaranteed properties of an ADC.

Check those assumptions against data and model behavior before interpreting a simulated schedule advantage. Useful diagnostics include:

Dose proportionality: compare dose-normalized intracellular profiles at more than one dose. Systematic changes in shape or peak per dose suggest saturation, dose-dependent access, or other nonlinear behavior.
Timing of observations: sparse samples near the predicted peak can make both its magnitude and timing poorly identified. Sampling should cover the expected lag, the rise, and the decline.
Residual patterns: consistent early or late prediction errors can indicate a missing transit step, an inaccurate release rate, or an omitted compartment rather than random noise.
Schedule consistency: if equal total doses produce different dose-normalized responses after accounting for timing, consider schedule-dependent target turnover, changing tumor access, or other mechanisms that violate the fixed-kernel assumption.

The linear kernel is useful for isolating how timing alone changes the profile. It is not a substitute for a model that represents processes such as saturable binding or uptake, changing antigen abundance, or dose-dependent linker stability when evidence supports them. A more detailed mechanistic model may predict changes in total AUC as well as peak concentration. Conversely, a simple comparison of AUCs can summarize total exposure but cannot show when the peak occurs or whether two profiles have different shapes.

Common pitfall: “The total dose is unchanged, so the peak and AUC must both be unchanged.” This confuses two different properties. In the stated linear model, complete AUC is unchanged because integrated responses add and each dose has the same kernel area. The peak can change because the responses overlap differently in time. The reverse error is also common: a lower peak does not by itself establish a lower full-course AUC.

Check Your Understanding

Try the questions first, then check each solution below. All values are hypothetical and use the kernel and assumptions defined in this tutorial.

Why does the model predict no instantaneous jump in active intracellular payload concentration at the time of dosing?
For two $50\ \mathrm{mg/m^2}$ doses at hours 0 and 10, calculate the model concentration at hour 10, immediately after the second dose.
Under the linear model, what is the complete AUC for a total dose of $80\ \mathrm{mg/m^2}$, regardless of whether it is given once or divided into two equal doses?
A higher dose produces a smaller-than-proportional increase in peak concentration. Which assumption should be questioned, and why?

Solutions. (1) The dose enters $S$, not $C$. It must pass through internalization and release before active payload appears, so $C$ changes continuously at the dosing time in this model. (2) The second dose has age zero and $h(0)=0$. The first dose has age 10 hours, so $C(10)=50h(10)+50h(0)$. Using $h(10)=2e^{-2.5}-7e^{-5}\approx0.1170$, $C(10)\approx50(0.1170)=5.85\ \mathrm{nM}$. (3) Since $F(\infty)=2$, the complete AUC is $80(2)=160\ \mathrm{nM\cdot h}$. Splitting the dose does not change this result if each dose has the same linear response. (4) Dose proportionality should be questioned. Saturable uptake, binding, or other concentration-dependent processes could cause the response per dose unit to fall as dose increases, invalidating simple superposition.

Dose Fractionation and Payload Release Timing in ADCs

What You'll Learn