Diagnosing ADC Response Differences With Matched Mechanistic Models

Why Match Exposure Before Comparing Mechanisms?

An ADC can produce a different cellular response because more conjugate reaches the tumor, because cells internalize it more efficiently, or because intracellular processing delivers more active payload. Payload release and efflux can further change the amount and timing of active drug inside a cell. If these mechanisms are allowed to vary together, a response difference alone rarely reveals which one is responsible.

A matched-exposure comparison holds the extracellular ADC concentration–time profile fixed across model variants, then changes one or more intracellular process parameters. This removes differences in the imposed ADC exposure as an explanation for predicted response differences. It does not make intracellular exposure equal: uptake and subsequent processing are precisely the mechanisms being compared. Ideally, the full concentration–time profile is matched; equal area under the curve alone may be insufficient when timing or nonlinear processes matter.

The earlier material on ADC species states, uptake, linker cleavage, intracellular payload release, and efflux motivates the states used here. The purpose now is to compare these mechanisms under a common input and ask which measurements could tell them apart.

A Minimal Model for Comparing ADC Processing

Consider three intracellular states, each measured in payload-equivalent amount per cell: \(A(t)\) is intact internalized ADC, \(L(t)\) is an intracellular cleaved payload-containing intermediate, and \(P(t)\) is active intracellular payload. Let \(C(t)\) be the matched extracellular ADC concentration in nM. A simple linear model is:

Here \(k_u\) is the uptake coefficient, \(k_{\mathrm{cl}}\) is the cleavage rate constant, \(k_{\mathrm{rel}}\) is the active-payload release rate constant, and \(k_{\mathrm{eff}}\) and \(k_{\mathrm{met}}\) describe payload efflux and intracellular loss, respectively. The yield \(y\) is the amount of active payload formed per amount of intermediate processed. The loss terms \(k_A\) and \(k_L\) represent competing removal of intact ADC and intermediate. All rate constants have units of day\(^{-1}\), except \(k_u\), which has units of fmol cell\(^{-1}\) day\(^{-1}\) nM\(^{-1}\); \(A,L,P\) are in fmol/cell.

This is a deliberately simplified model. It assumes well-mixed states, first-order processing and loss, a known shared \(C(t)\), and fixed parameters during the comparison. It also treats \(y\) as constant and does not explicitly represent binding, DAR subpopulations, or cell-to-cell variability. Those omissions matter: the model is useful for structured comparisons, not a claim that every ADC follows these exact kinetics.

For a response link, one possible assumption is that active payload produces a concentration-dependent killing hazard \(h(P)\), so viable cell number \(N(t)\) satisfies \(dN/dt=-h(P)N\). For example, \(h(P)=h_{\max}P/(EC_{50}+P)\), where \(h_{\max}\) is a maximum hazard and \(EC_{50}\) is the payload amount per cell giving half that hazard. This makes clear why intracellular payload time courses are mechanistically informative: they are upstream of the modeled killing response.

What Steady State Can and Cannot Show

If \(C(t)=C\) is held constant long enough for the states to approach equilibrium, setting each derivative to zero gives:

These equations give a direct comparison of the predicted intracellular states under a common constant ADC input. They also show why a terminal payload measurement can be ambiguous: several rate constants enter the same steady-state expression. A change in \(P^*\) does not uniquely identify which parameter changed.

Time Courses Help Separate Processing Steps

The steady-state comparison discards timing. To see why timing matters, assume the initial intracellular states are zero and take Laplace transforms of the linear equations. Successive substitution gives the transfer function from extracellular ADC concentration to active payload:

Each denominator factor represents a state’s processing and loss timescale. Changing cleavage, release, or efflux can therefore alter the rise and decline of payload even when the exposure profile is unchanged. But the numerator contains a product of parameters, so a single output curve may still leave trade-offs: for example, a higher uptake coefficient and a lower yield can produce similar payload input.

For a constant input switched on at time zero, the early-time expansion also illustrates the sequence of steps. With zero initial states, \(A\) first grows approximately as \(k_uCt\), then \(L\) as \(k_{\mathrm{cl}}k_uCt^2/2\), and then \(P\) as:

This follows by integrating each successive production term once, while higher-order loss effects are negligible at very small \(t\). It predicts a lagged payload rise, but the early rise depends on a product of rates; it does not independently identify uptake, cleavage, and release.

Efflux Can Reverse the Interpretation of an Extracellular Measurement

Efflux removes active intracellular payload but transfers material out of the cell. Thus, high extracellular payload need not mean high intracellular payload, and measuring only one compartment can be misleading. The model predicts the efflux flux as \(k_{\mathrm{eff}}P\), in payload amount per cell per day. If exported payload is stable and recovered quantitatively, its time course provides information complementary to intracellular \(P\).

Choosing Measurements and Diagnosing Model Failure

A useful matched-exposure experiment should measure the shared input and several points along the proposed mechanism, not only the final response. Possible measurements include extracellular ADC concentration, surface-associated versus internalized ADC, intracellular intact conjugate, cleaved intermediate, active intracellular payload, extracellular payload, and viable cell number. Measurements should span early uptake, intermediate processing, payload accumulation or washout, and the later response. The relevant times depend on the model’s predicted timescales; sampling only after all transients have passed sacrifices kinetic information.

Check that the exposure profiles are genuinely comparable and that the assay distinguishes the chemical species assigned to each model state. A measured response difference can also reflect differences in cell number at baseline, antigen abundance, or payload sensitivity, so these should be controlled or represented explicitly. A poor fit to both intracellular species and response is a warning that the chosen states or kinetics may be inadequate, not merely that a parameter needs adjustment.

Several diagnostic patterns are useful, but none is conclusive in isolation:

• If internalized ADC is already mispredicted before payload appears, investigate uptake measurement, the imposed input, or the intact-state model before attributing the discrepancy to release.
• If intact ADC is described but the cleaved intermediate is not, cleavage or competing intermediate formation and loss may be misspecified.
• If intracellular payload is overpredicted while measured exported payload is high, an efflux or extracellular recovery process may be missing or incorrect.
• If all species are described but cell killing is not, payload-to-response assumptions, cell sensitivity, or response heterogeneity may be inadequate.

These comparisons differ from an empirical exposure–response model, which can summarize a relationship between exposure and response without assigning the difference to an intracellular step. They also improve on comparing a single endpoint across candidate mechanisms, because species-specific time courses constrain the sequence of intermediate states. However, a more detailed mechanistic model is not automatically more identifiable or more predictive: every added process requires suitable data or defensible prior information.

Common Pitfall and Practical Limits

Parameter identification can remain weak even with good fits. In the transfer function, uptake and yield both scale payload production, while processing rates affect overlapping time patterns. If only payload and viability are observed, distinct parameter combinations may produce similar outputs. A profile of the cleaved intermediate, for example, can help separate a cleavage change from a downstream release change. In practice, assess whether the candidate models make meaningfully different predictions at available sampling times and whether measurement uncertainty is small enough to resolve those differences.

Finally, “matched exposure” is a modeling condition, not a universal biological guarantee. If only total ADC is matched while conjugated species or DAR distributions differ, the payload-equivalent input may not be matched. If the exposure profile is nonlinear or time-varying, the constant-input steady-state formulas above do not describe the full trajectory. Such limitations should be stated rather than hidden by a precise-looking response prediction.