Maxwell Chirehwa, Alienor Berges, David Fairman, Iain Uings, Núria Buil-Bruna 1
1 GSK (, UK)
Objectives
Interleukin-18 (IL-18) is a pro-inflammatory cytokine that amplifies innate and adaptive immune responses. In circulation, IL-18 is tightly regulated by high-affinity binding to IL-18 binding protein (IL-18BP), an endogenous soluble decoy receptor that neutralises IL-18 and prolongs its apparent half-life. GSK1070806 (aletekitug) is a monoclonal antibody targeting IL-18 that binds to both free IL-18 and IL-18 bound to IL-18BP. The presence of IL-18BP complicates quantitative characterisation of IL-18 and aletekitug pharmacokinetics and pharmacodynamics.
The objective of this work was to develop mechanistically informed and identifiable PK/PD models to describe IL-18 dynamics, quantify target engagement, and support dose and schedule selection for aletekitug.
Methods
Model development proceeded sequentially.
Model 1 was a mechanistic buffering model constructed using literature and in vitro binding data following established target-mediated drug disposition (TMDD) principles [1,2]. It explicitly represented IL-18, IL-18BP, antibody, and binary and ternary complexes using mass-action kinetics and pool-specific clearance. The model described IL-18 stabilisation through redistribution into IL-18BP-bound and antibody-bound pools. Clinical data, which showed rapid sustained formation of circulating IL-18-drug complexes until the last measurable timepoint [3], was used only for qualitative assessment of physiological plausibility. The lack of observable decline in total IL-18 prevented reliable estimation of the endogenous IL-18 degradation rate and in vivo binding affinity (KD).
A subsequent clinical study (NCT05590338) was designed using lower doses and/or extended follow-up, allowing observation of IL-18 recovery. These data enabled estimation of IL-18 turnover and KD. Clinical observations included aletekitug and total IL-18 levels over time. Free IL-18 was not measured, consistent with best practices for soluble target assays [4]. The full mechanistic model was fitted to these data to estimate system-specific and drug-specific parameters.
Two reduced model were then derived:
Model 2 applied a quasi-equilibrium approximation [1,2], assuming rapid binding between IL-18, IL-18BP, and antibody. Ordinary differential equations (ODE)s for the IL-18:IL-18BP, IL-18:antibody and IL-18:IL-18BP:antibody complexes were removed, while total IL-18 and total IL-18BP remained dynamic states. Free IL-18 and fractional occupancy were reconstructed algebraically, and clearance was modelled using pool-weighted elimination.
Model 3 was a parsimonious turnover model, collapsing IL-18BP dynamics and representing total IL-18 as a single turnover species with drug-dependent elimination. Endogenous buffering influenced target engagement through fractional occupancy but was assumed to not dynamically modify IL-18 clearance.
Models were compared based on structural assumptions, parameter identifiability, computational stability, and their ability to reproduce total IL-18 dynamics and predict target engagement across dosing regimens.
Results
The full TMDD mechanistic buffering model reproduced drug-induced IL-18 accumulation and provided biological interpretability but required many parameters and was poorly identifiable with the available clinical data.
The quasi-equilibrium model retained the key buffering mechanism with fewer parameters and reproduced total IL-18 dynamics similarly to the mechanistic model, although results depended on assumptions about IL-18BP turnover.
The parsimonious turnover model enabled stable estimation of KD and IL-18 degradation from longitudinal data collected under lower exposure and extended follow-up. While buffering effects were represented indirectly, this model was robust and computationally efficient and was deemed fit-for-purpose for dose exploration.
Predicted target engagement was consistent across models when binding affinities were aligned.
Conclusions
Modelling IL-18 in the presence of endogenous IL-18BP requires explicit consideration of buffering and stabilisation mechanisms. A mechanistic TMDD framework informed by literature and in vitro data provided foundational system understanding but was limited by parameter identifiability. Reduced models based on rapid-binding and turnover approximations preserved target engagement while improving robustness and parsimony when supported by appropriate data. These results highlight the importance of aligning clinical study design with modelling objectives and support a staged strategy combining mechanistic insight, targeted data generation, and principled simplification to guide dose selection for cytokine-targeting therapies.
References:
1. Gibiansky L, Gibiansky E. Target-mediated drug disposition model: relationships between pharmacokinetic parameters and drug-target binding kinetics. J Pharmacokinet Pharmacodyn. 2009;36:497–516.
2. Mager DE, Jusko WJ. General pharmacokinetic model for drugs exhibiting target-mediated drug disposition. J Pharmacokinet Pharmacodyn. 2001;28:507–532.
3. Mistry P, Reid J, Pouliquen I, et al. Safety, tolerability, pharmacokinetics, and pharmacodynamics of single-dose anti-interleukin-18 mAb GSK1070806 in healthy and obese subjects. Int J Clin Pharmacol Ther. 2014;52:867–879.
4. Fairman D, Tang H. Best practices in mAb and soluble target assay selection for quantitative modelling and qualitative interpretation. AAPS J. 2023;25:23.
Reference: PAGE 34 (2026) Abstr 12087 [www.page-meeting.org/?abstract=12087]
Poster: Drug/Disease Modelling - Other Topics