QSP Models of Antibody-Drug Conjugates

1. What Is an Antibody-Drug Conjugate?

An antibody-drug conjugate (ADC) is a targeted therapeutic constructed from an antibody, a linker, and one or more potent drug molecules, commonly called payloads or warheads. The antibody provides target recognition and delivery, while the linker controls how the payload remains attached to the antibody and how it is released.

An ADC is therefore not simply an antibody carrying a fixed amount of drug. After administration, multiple molecular species can coexist, including intact conjugated antibody, partially deconjugated species, unconjugated antibody, and released payload. The drug-to-antibody ratio (DAR) can also change over time because of deconjugation.

Published ADC QSP models have incorporated mechanisms at cellular, tumor, preclinical, and clinical scales, with applications ranging from understanding ADC distribution and payload release to translating preclinical efficacy into human predictions.

2. Why Do ADCs Need QSP Models?

Traditional PK models can describe observed concentrations efficiently, but ADCs create a particularly complex modeling problem because several related molecular species have different properties and because the drug changes as it moves through the body.

These features motivate a mechanistic framework in which the ADC, its components, the biological system, and the disease state are represented explicitly. Reviews of ADC systems pharmacology describe models spanning cellular mechanisms, spatial tumor representation, preclinical translation, and clinical translation.

3. The Architecture of an ADC QSP Model

A useful ADC QSP model can be viewed as a series of connected biological layers. The exact implementation varies by molecule and scientific question, but a common conceptual architecture is:

Each arrow represents one or more mechanistic processes. A model may include only the processes necessary for the intended application, or it may expand into a multiscale platform model.

4. The Multiple Analytes Problem

One of the defining challenges in ADC modeling is that there is no single concentration measurement that completely describes the pharmacology of the molecule. Different bioanalytical assays can measure different aspects of the ADC system.

The choice of analytes is therefore part of the model-development strategy. Published ADC QSP work has emphasized measuring conjugated antibody or antibody-conjugated drug together with total antibody and unconjugated drug when possible, because these measurements help separate the major processes represented by the model.

5. Drug-to-Antibody Ratio: Why DAR Matters

The drug-to-antibody ratio (DAR) is the average number of payload molecules attached to an antibody molecule. ADC products can contain a distribution of DAR values rather than a single uniform DAR.

A simplified representation is:

where \(ADC_{DAR=j}\) represents the concentration of ADC molecules carrying \(j\) payload molecules.

In a simple mixture, the average DAR can be written as:

The average DAR is useful, but it can hide important information. Two ADC products could have the same average DAR while having different distributions of low-, medium-, and high-loaded species.

A mechanistic model can represent these species explicitly when the data and scientific question justify that level of detail.

6. Modeling Deconjugation and Payload Release

The DAR of an ADC can change after administration because payload molecules may be released from the antibody. This process is commonly called deconjugation.

A simple sequential model can represent a high-DAR species converting into a lower-DAR species:

where \(A_j\) is the amount of ADC with DAR \(j\), and \(k_{dec,j}\) is a deconjugation rate associated with that species.

A released payload may then enter a separate systemic payload compartment:

This is intentionally simplified. In a real ADC model, payload release may occur through several pathways, including systemic deconjugation and intracellular processing after target-mediated uptake.

Recent reviews emphasize that linker chemistry can influence payload release kinetics and therefore the pharmacokinetic behavior of ADCs.

7. Target Binding and Target-Mediated Drug Disposition

Many ADCs are designed to bind a tumor-associated antigen. The interaction between ADC and target can therefore affect both pharmacology and disposition.

A simple binding reaction is:

where \(R\) denotes the target receptor or antigen.

The corresponding binding dynamics can be represented as:

If the ADC-target complex is internalized, the model can include an additional process:

This creates a mechanistic connection between extracellular exposure and cellular uptake.

Target-mediated drug disposition (TMDD) can become important when target binding and internalization contribute materially to ADC clearance or tissue distribution. Mechanistic ADC TMDD models have been developed specifically to account for distribution, target binding, toxin load, deconjugation, and released-toxin PK.

8. Internalization and Intracellular Processing

After binding to the target, an ADC may be internalized by the target cell. A QSP model can represent this process using intracellular compartments.

For a simplified model, the intracellular payload concentration \(P_i\) might be represented as:

where \(A_{int}\) represents an internalized ADC-related amount and \(k_{rel}\) describes the effective payload-release process.

More detailed QSP models can distinguish endosomal and lysosomal compartments, receptor recycling, linker-dependent payload release, and intracellular drug disposition.

9. Bystander Killing

Some ADC payloads can cross cell membranes after release. As a result, payload generated inside a target-positive cell may affect neighboring target-negative cells. This phenomenon is commonly called a bystander effect.

A simplified model might represent intracellular payload movement between two cell populations:

where \(P_{+}\) represents payload in target-positive cells and \(P_{-}\) represents payload exposure in neighboring target-negative cells.

Bystander effects can therefore change how antigen heterogeneity translates into tumor response. A model that assumes every tumor cell must independently bind ADC may miss this mechanism.

10. Tumor Penetration and Spatial Heterogeneity

Tumors are not homogeneous compartments. Blood vessels, extracellular space, cell density, antigen expression, interstitial transport, and binding can all vary throughout a tumor.

A QSP model can represent tumor distribution using spatial compartments or continuous spatial models. A simplified spatial representation might divide a tumor into vascular, peripheral, and deeper regions.

A particularly important phenomenon is the binding-site barrier. Very strong binding can increase uptake near the tumor vasculature while limiting penetration farther into the tumor. Mechanistic modeling has been used to study this tradeoff quantitatively.

Thus, greater target affinity does not necessarily imply greater payload delivery to every region of a tumor. The relationship depends on transport, target density, internalization, and the spatial distribution of ADC.

11. Connecting Intracellular Payload to Cell Killing

Once intracellular payload concentration has been predicted, a QSP model needs a pharmacodynamic relationship that connects payload exposure to cellular effect.

A simple \(E_{\max}\)-type model can be written as:

For cytotoxic payloads, however, a cell-kill formulation may be more useful. One possible representation is:

where \(N\) is the viable tumor-cell population and \(k_{kill}(P)\) increases with intracellular payload concentration.

For example:

Here \(h\) is a Hill coefficient describing the steepness of the concentration-effect relationship.

12. Linking ADC Exposure to Tumor Growth and Response

A mechanistic ADC QSP model can connect predicted payload exposure to tumor growth. A simple tumor-growth model is:

where \(T\) represents viable tumor burden.

The model can then simulate tumor growth in the absence of treatment and tumor suppression following ADC administration.

Published ADC QSP models have connected tumor payload concentrations to tumor-growth inhibition in preclinical systems and, in some cases, translated these relationships to human clinical outcomes.

13. Separating Drug Parameters from System Parameters

A powerful feature of QSP modeling is the explicit distinction between properties of the drug and properties of the biological system.

This distinction matters for translation. A drug-specific parameter may remain approximately associated with the molecule when the model is moved between systems, while a biological parameter may need to change between species, tumor models, or patient populations.

For example, target expression and tumor growth rate are system characteristics, whereas ADC affinity and payload potency are primarily properties of the drug. Keeping these concepts separate helps prevent inappropriate transfer of parameters from one biological context to another.

14. Translating an ADC QSP Model from Animals to Humans

One of the major applications of ADC QSP modeling is connecting preclinical experiments to human predictions.

A conceptual translation workflow is:

The model can use preclinical information about antibody disposition, target expression, cellular uptake, payload release, and tumor response while replacing species-specific physiological quantities with human values.

Physiologically based or semi-mechanistic components can help describe tissue distribution, while cellular QSP components can preserve mechanistic relationships between target binding, internalization, payload exposure, and effect.

A published platform QSP model for ADCs demonstrated this type of multiscale approach using data from trastuzumab emtansine and trastuzumab deruxtecan, including ADC disposition, tumor uptake, intracellular processing, payload release, and tumor-growth response.

15. Modeling the Therapeutic Window

An ADC is designed to increase the separation between tumor exposure and unwanted normal-tissue exposure. A QSP framework can therefore be used to examine not only efficacy but also potential toxicity mechanisms.

A conceptual therapeutic-index model can compare:

This is not a universal quantitative definition of therapeutic index, but it illustrates the modeling objective: understand how ADC design and biological context jointly determine efficacy and toxicity.

Potential toxicity mechanisms can include:

• Normal-tissue expression of the target antigen.
• Non-target uptake of ADC.
• Systemic release of free payload.
• Payload distribution to sensitive organs.
• Off-target intracellular processing.
• Accumulation associated with repeated dosing.

Compared with efficacy, ADC toxicity remains a less developed area of QSP modeling, and translating mechanistic toxicity predictions into humans can require substantial additional data and model qualification.

16. Using QSP to Explore ADC Design Choices

One of the most useful features of an ADC QSP model is the ability to vary drug design parameters systematically.

The model can be used to simulate combinations of these variables rather than testing every combination experimentally.

For example, simulations can compare a highly potent payload with strong bystander activity against a less permeable payload, while holding antibody targeting characteristics constant.

This does not mean that the model determines the optimal ADC automatically. Instead, it provides a quantitative framework for exploring hypotheses and identifying design regions that warrant experimental investigation.

17. Worked Example: A Minimal ADC QSP Model

Consider a hypothetical ADC administered intravenously. Assume the model contains four main components:

1. Systemic ADC concentration \(A\).
2. Tumor-bound ADC \(B\).
3. Intracellular payload concentration \(P\).
4. Viable tumor burden \(T\).

Step 1: Systemic ADC disposition

Assume a simple first-order systemic clearance:

For illustration, suppose:

• \(CL_{ADC}=0.20\) L/h
• \(V_{ADC}=3.0\) L
• \(k_{on}=0.02\) L/(nmol·h)
• \(k_{off}=0.10\) h\(^{-1}\)

The systemic elimination rate constant in the simple linear component is:

Step 2: Target binding

Suppose the tumor target concentration is initially \(R=100\) nmol/L and binding is represented by:

If \(k_{int}=0.30\) h\(^{-1}\), the bound ADC is not simply an equilibrium pool: it is also an input into intracellular processing.

Step 3: Intracellular payload generation

Assume an effective payload-release rate \(k_{rel}=0.15\) h\(^{-1}\):

Suppose \(k_{loss}=0.25\) h\(^{-1}\). The intracellular payload is therefore driven by target-mediated ADC uptake and removed according to the assumed intracellular loss processes.

Step 4: Payload-driven tumor-cell killing

Let the payload-dependent killing rate follow a Hill relationship:

Suppose:

• \(k_{max}=0.05\) h\(^{-1}\)
• \(EC_{50}=2\) arbitrary payload concentration units

If the model predicts \(P=2\), then:

Step 5: Tumor dynamics

Assume a tumor-growth rate of \(k_g=0.015\) h\(^{-1}\):

Under these illustrative conditions, the instantaneous model predicts net tumor regression because the modeled killing rate exceeds the modeled growth rate.

18. Which ADC Parameters Matter Most?

A mechanistic model may contain dozens or hundreds of parameters. Sensitivity analysis helps identify which parameters have the greatest influence on a selected model output.

For a parameter \(\theta\) and output \(Y\), a local normalized sensitivity can be expressed conceptually as:

Possible outputs include:

• Tumor payload concentration.
• Maximum tumor-growth inhibition.
• Tumor exposure over a treatment cycle.
• Free-payload exposure in plasma.
• Normal-tissue payload exposure.
• Time to tumor progression in a simulated system.

Sensitivity analysis can reveal whether a model prediction is primarily controlled by target expression, binding affinity, internalization, deconjugation, payload potency, tumor growth, or another process.

This can be especially useful for deciding which experimental measurements would most reduce uncertainty.

19. Parameter Uncertainty and Model Uncertainty

QSP predictions should not be interpreted as exact because both parameters and model structure are uncertain.

A useful QSP analysis therefore examines not only a nominal simulation but also how predictions change when uncertain inputs are varied.

20. What Is an ADC Platform QSP Model?

A platform QSP model is designed to provide a reusable mechanistic framework rather than a model built exclusively for one ADC.

The platform can contain common biological components such as:

• Systemic antibody disposition.
• Tumor and normal-tissue distribution.
• Target expression and turnover.
• ADC binding and internalization.
• Intracellular trafficking.
• Linker-dependent payload release.
• Payload distribution and loss.
• Bystander effects.
• Tumor growth and cell killing.

Drug-specific parameters can then be replaced or calibrated for different ADCs. This creates a distinction between the reusable biological framework and the properties specific to an individual therapeutic.

Recent platform work has explored this approach using different ADCs sharing the same antibody target but differing in linker-payload architecture and intracellular mechanisms.

21. A Practical Workflow for Building an ADC QSP Model

1. Define the scientific question. Determine whether the model is intended for candidate selection, mechanism exploration, efficacy prediction, toxicity analysis, dose selection, or clinical translation.
2. Define the ADC components. Characterize the antibody, target, linker, payload, DAR distribution, and relevant release mechanisms.
3. Map the analytes. Identify which experimental measurements correspond to total antibody, conjugated ADC, free payload, DAR species, target occupancy, biomarkers, and tumor response.
4. Build the systemic PK layer. Represent the disposition of the ADC and relevant released species.
5. Add target-mediated processes. Represent target binding, internalization, recycling, and degradation when relevant.
6. Add intracellular processing. Include payload release and intracellular disposition at the level necessary for the scientific question.
7. Add tumor biology. Represent tumor growth, target expression, cellular sensitivity, and relevant heterogeneity.
8. Connect exposure to response. Use an appropriate pharmacodynamic or tumor-growth model.
9. Calibrate and qualify. Use multiple independent data types where possible and test whether the model can reproduce observations not used directly for calibration.
10. Perform sensitivity and uncertainty analysis. Identify influential parameters and determine how uncertainty propagates to important predictions.
11. Translate across scales. Distinguish drug-specific parameters from species- or system-specific parameters.
12. Use the model for simulation. Explore dosing, ADC design, target expression, payload properties, and other scenarios that would be difficult to evaluate experimentally.

22. How Should an ADC QSP Model Be Qualified?

Model qualification asks whether the model is adequate for its intended use. A model does not need to reproduce every experimental observation perfectly to be useful, but its limitations should be understood.

Useful qualification activities include:

• Comparison of predicted and observed concentration-time profiles.
• Evaluation of ADC and free-payload exposure.
• Assessment of target-binding and internalization behavior.
• Reproduction of independent tumor-growth experiments.
• Evaluation across different doses or schedules.
• Testing across different tumor models when appropriate.
• Sensitivity analysis of key parameters.
• External validation using data not used during model development.

For translational models, qualification should also examine whether the model can reproduce known preclinical-to-clinical relationships before it is used for prospective prediction.

23. What ADC QSP Models Do Not Tell Us Automatically

Mechanistic detail does not eliminate uncertainty. Several limitations should remain explicit when interpreting ADC QSP simulations.

• A mechanistic model is not a complete representation of biology. Important processes may be omitted for tractability.
• Parameter identifiability can be difficult. Multiple parameters may produce similar predictions when data are sparse.
• Average target expression may hide spatial heterogeneity. Two tumors with the same average expression can behave differently.
• In vitro potency does not automatically determine in vivo efficacy. Exposure, penetration, intracellular processing, and tumor biology intervene.
• Preclinical tumor models are imperfect representations of human disease. Differences in tumor microenvironment, target expression, immune biology, and growth kinetics can affect translation.
• Free payload is not necessarily the only driver of efficacy. The relevant pharmacologic driver depends on the payload and mechanism of action.
• Predictions are conditional on assumptions. Changing the structural model or parameter assumptions can change the prediction.
• Toxicity prediction remains challenging. Normal-tissue sinks, off-target exposure, and organ-specific mechanisms may require additional biological detail.

24. What Can ADC QSP Models Be Used For?

Once appropriately developed and qualified, ADC QSP models can support a range of questions across the drug-development lifecycle.

25. Where Does ADC QSP Fit Among Other Models?

ADC development can use several complementary modeling approaches. They should not be viewed as mutually exclusive.

The appropriate model depends on the scientific question. A complex QSP model should not replace a simpler model when the simpler model adequately answers the question.

26. Where Are ADC QSP Models Going?

ADC QSP modeling is moving toward increasingly integrated multiscale frameworks. Several areas are particularly important.

• More explicit tumor heterogeneity: models can represent distributions of target expression rather than a single average tumor value.
• More detailed intracellular biology: endosomal trafficking, lysosomal processing, linker chemistry, and payload release can be represented with greater mechanistic resolution.
• Improved toxicity models: normal-tissue sinks and organ-specific payload exposure can be integrated with efficacy models.
• Platform models: common biological structures can be reused across ADCs with different targets, linkers, and payloads.
• Virtual populations: variation in target expression, tumor burden, growth, and other patient factors can be incorporated into simulations.
• Model-informed ADC design: QSP simulations can be used earlier to compare combinations of antibody, linker, payload, and DAR characteristics.

The long-term objective is a model that can preserve enough mechanistic detail to explain why an ADC behaves differently across targets, tumors, patients, and design architectures while remaining identifiable and useful for development decisions.

Where to Go Next

A natural next step is to study the individual mechanisms that make ADC QSP models different from conventional PK models: drug-to-antibody ratio and deconjugation, target-mediated drug disposition, tumor penetration and the binding-site barrier, intracellular payload release, and bystander effects.

These concepts can then be combined into a multiscale ADC model that links systemic PK to tumor exposure and ultimately to tumor-growth inhibition. From there, the model can be extended to preclinical-to-clinical translation, virtual populations, and therapeutic-window analysis.

References

1. Lam I, Reddy VP, Ball K, et al. Development of and insights from systems pharmacology models of antibody-drug conjugates. CPT: Pharmacometrics & Systems Pharmacology. 2022;11:967–990. https://pmc.ncbi.nlm.nih.gov/articles/PMC9381915/
2. Khera E, et al. Towards a platform quantitative systems pharmacology (QSP) model for preclinical to clinical translation of antibody drug conjugates (ADCs). 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC11576657/
3. Pharmacokinetic and Pharmacodynamic Modeling of Antibody-Drug Conjugates. Review of empirical, semi-mechanistic, PBPK, cellular, and systems-level ADC PK/PD models. https://pmc.ncbi.nlm.nih.gov/articles/PMC12785113/
4. Singh AP, et al. Application of Pharmacokinetic-Pharmacodynamic Modeling and Simulation for Antibody-Drug Conjugate Development. Pharmaceutical Research. 2015;32:3508–3525. https://pubmed.ncbi.nlm.nih.gov/25666843/
5. Target-mediated drug disposition model and its approximations for antibody-drug conjugates. Mechanistic treatment of ADC target binding, disposition, deconjugation, and released toxin. https://pubmed.ncbi.nlm.nih.gov/24322877/
6. Impact of Physiologically Based Pharmacokinetics, Population Pharmacokinetics and Pharmacokinetics/Pharmacodynamics in the Development of Antibody-Drug Conjugates. Review of quantitative modeling approaches used across ADC development. https://pmc.ncbi.nlm.nih.gov/articles/PMC7756373/
7. Pharmacokinetic modeling frameworks for complex antibody-drug conjugates analytes: clinical pharmacology considerations. Review of ADC analyte heterogeneity, DAR distributions, deconjugation, TMDD, and mechanistic PK modeling. https://pubmed.ncbi.nlm.nih.gov/42604672/