Computational Systems Pharmacology of Pyrrolobenzodiazepine-Based Antibody-Drug Conjugates

Introduction:
Antibody-Drug Conjugates (ADCs) are engineered immunoconjugate drugs comprising 3 core components: (1) a monoclonal antibody (Ab) and (2) one or more cytotoxic small molecules (known as warheads), attached via (3) a chemical linker. The Ab targets a specific tumor antigen, so the warhead can selectively kill cancer cells while sparing healthy tissue. As a designed, targeted drug delivery vehicle to selected cells, one might expect ADCs to have high success rates; however, despite hundreds of clinical trials, only 14 have been approved for use in oncology, and those that are approved only succeed in a fraction of patients. Seeking answers through experimental methods alone can be laborious, expensive, or even impossible. Quantitative systems pharmacology (QSP) approaches integrate mechanistic knowledge with biomedical data at multiple scales to construct interpretable, predictive computational models, which can help us explore mechanisms and hypotheses efficiently, and support decision-making in drug development and clinical practice. Strengthening our understanding of ADCs through QSP modeling will support the development of improved cancer treatments.



Materials and Methods:
We have expanded upon our computational systems pharmacology model of cellular and intracellular mechanisms specific to ADCs with pyrrolobenzodiazepine (PBD) warheads, developing a compartmental model to predict in vivo efficacy and incorporating the pharmacokinetics and pharmacodynamics (PK/PD) of the ADCs in mice. The model is composed of coupled, deterministic, nonlinear, ordinary differential equations (ODEs), outputting the concentrations of each molecule or molecular complex over time. Model parameters were estimated from literature, or calibrated and validated using experimental data on in vitro cytotoxicity and in vivo tumor growth inhibition provided by AstraZeneca for anti-BCMA and anti-HER2-PBD ADCs; the antibodies carry the same payloads (the attached linkers and warheads) but target different antigens for hematological (BCMA) and solid (HER2) tumors respectively. The model describes ADC binding, internalization, recycling, intracellular trafficking and degradation, efflux and influx of warhead, warhead binding, DNA crosslink formation, cell killing, and systemic distribution and clearance.



Results, Conclusions, and Discussions:
To translate the cellular mechanistic model to incorporate in vivo murine PK/PD, we added compartments to reflect the system depending on the tumor type targeted by each ADC; hematological tumors are represented with 2 compartments (central and peripheral), while solid tumors are represented with 3 compartments (central, peripheral, and tumor). Following calibration and parameterization of the model, we conducted local sensitivity analyses at different time points and receptor expression levels to determine the effect on the tumor cell population. At both the in vitro and in vivo scales, the system was most sensitive to parameters for cell killing and DNA crosslinking, as well as ADC uptake, endosomal exit, and recycling, suggesting that both warhead potency and non-specific ADC uptake are important considerations that influence the efficacy of these ADCs. Using this multi-scale model, we simulated ADC efficacy under various ADC design scenarios and examined effects specific to both the anti-BCMA and anti-HER2 ADCs at the in vitro and in vivo scales. Notably, we built a warhead tracking feature into the model, enabling us to visualize changes in PBD warhead location and recent history at multiple scales. Tracking levels of warhead at the cellular and compartmental levels revealed additional insights into warhead movement and mechanisms of action. For instance, the model showed that the PBD-based ADCs, despite having noncleavable linkers that only release warhead intracellularly, still demonstrate a significant amount of bystander effects due to the substantial amount of warhead influxed following efflux from other cells.

In summary, we have developed a multi-scale, mechanistic QSP model describing the efficacy of PBD-ADCs to enable in vitro to in vivo translation. This model integrates cellular-level mechanisms specific to PBD-based ADCs with in vivo mouse PK/PD, using parameters from literature or fit to AZ experimental data. By incorporating relevant experimental data, this computational model can be further developed to create a human clinical model to run virtual clinical trials.



References:
1. Coats, S., et al., Antibody–Drug Conjugates: Future Directions in Clinical and Translational Strategies to Improve the Therapeutic Index. Clinical Cancer Research, 2019. 25(18): p. 5441-5448.
2. Xie, L., E.J. Draizen, and P.E. Bourne, Harnessing big data for systems pharmacology. Annual review of pharmacology and toxicology, 2017. 57: p. 245-262.