Daniela Santurio 1, Aiman Nazki 1, Sirin Yonucu 1, Birses Debir 1, Maciej Swat 1, Rachel Rose 1, Piet van der Graaf 1,2, Andrzej Kierzek 1,3
1 Certara (Sheffield, United Kingdom), 2 Universiteit Leiden (Leiden, The Netherlands), 3 University of Surrey (, United Kingdom)
Introduction: Over last decade the Quantitative systems pharmacology (QSP) has matured to an approach recognized by regulatory agencies, which is reflected by the growing number of submissions where QSP models were used as evidence to inform drug development [1]. In this work, we present – IO Simulator – a QSP modelling platform for solid tumours, which incorporates the action and the underlying biology of immuno-oncology (IO) drugs. The core of IO Simulator is a mechanistic framework that characterizes how these drugs modulate the Cancer Immunity Cycle [2], complex and dynamic biological process in the tumour–immune system. To predict potential stratification and efficacy biomarkers, seven cellular and eight cytokine species were included into the model. Moreover, IO Simulator was developed modularly, which enables predictive modelling of combination therapies by incorporating multiple targets and pathways, each integrated at its own level of granularity. Finally, we demonstrate the application of IO Simulator to small-cell lung cancer (SCLC) by evaluating Tarlatamab, a DLL3-targeted bispecific T-cell engager (BiTE), together with other immunotherapeutic agents such as Atezolizumab (a PD-L1 inhibitor), Nivolumab (a PD-1 inhibitor), and Ipilimumab (a CTLA-4 inhibitor).
Objectives:
• Develop a mechanistic SCLC virtual population (VPop) with model IO Simulator using clinical data from Tarlatamab and Atezolizumab, capturing key inter‑patient variability.
• Demonstrate how IO Simulator‑based VPop predicts monotherapy and combination therapy efficacy, biomarkers, and likelihood of cytokine release syndrome (CRS).
• Evaluate the predictive performance and generalizability of model IO Simulator and the VPop using independent datasets, such as Nivolumab and Ipilimumab.
Methods: The main IO Simulator workspace simulates the Cancer Immunity Cycle in solid tumours across tumour microenvironment, lymph nodes, and blood, including immune cells and cytokines. For cytokine dynamics, a minimal biologics PBPK model based on Simcyp Simulator [3], was calibrated with therapeutic cytokine data, then linked to the IO Simulator by replacing therapeutic dosing with endogenous production of cytokines in cancer immunity cycle model and adding a new TME compartment.
To add a new drug modality to IO Simulator, we first developed dedicated mechanism of action (MoA) models: BiTE model was based on Betts et al. [4] and bespoke models for anti-PD-1/PD-L1/CTLA-4 IO agents. Calibrated with in vitro data, these linked to a minimal biologics PBPK model [3] with tumour disposition, calibrated using in vivo data [6,7]. Finally, the combined MoA/PBPK models were connected to the main IO Simulator workspace.
To capture biological variability of individual tumour and patient responses, we developed a Virtual population (VPop) using established approach of Allen et al. [5]. The VPop was calibrated using efficacy and cytokine data from Tarlatamab [6] and Atezolizumab [7] clinical trials, SCLC composition from deconvoluted cell fractions in omics datasets [8] and other SCLC-intrinsic parameters—such as tumour growth rates and DLL3 expression levels—sourced from published literature [9].
Following virtual population calibration, we assessed CRS risk using an incidence-calibrated IL-6 threshold derived from Tarlatamab aggregate clinical summaries [7]. Individual peak IL-6 concentrations were assumed to follow a log-normal distribution, with parameters estimated from reported statistics (median, IQR, mean or SD). We further evaluated the predictive performance and generalizability of the model using efficacy data from independent datasets such as Nivolumab and Ipilimumab [10].
Results: An IO Simulator‑based VPop with 100 patients that statistically reproduce the SCLC clinical population was generated. This calibrated VPop accurately recapitulated clinical efficacy (OR~50%,48% Tarlatamab (10mg, 100mg), ~12% Atezolizumab (1200mg)), while capturing tumour heterogeneity and cytokine dynamics. Incidence-calibrated IL-6 thresholds stratified patients into CRS-likely vs CRS-unlikely categories across Tarlatamab regimens: 10 mg first dose (41% observed vs 43% predicted), 10 mg second dose (30% vs 33%), 100 mg first dose (29% vs 43%), and 100 mg second dose (49% vs 17%) – indicating that predictive performance could be improved with patient-specific data of greater granularity. Tarlatamab–Atezolizumab combination simulations predicted 9% OR improvement vs Tarlatamab monotherapy. The model and VPop generalized well to independent Nivolumab/Ipilimumab datasets, predicting efficacy for all dosing protocols.
Conclusion: The context of use of this case study is the application of a QSP platform model to inform decisions about dose adjustments at late Phase I or at Phase II stages of drug development. IO Simulator was calibrated with existing SCLC pre-clinical and clinical data, and our findings suggest that it can be applied to predict – monotherapy or combination therapy -efficacy and biomarkers for the next stage of clinical trial. Using a prediction-validation framework, IO Simulator provided reasonable evidence for possible adjustments of dosing regimen to achieve target efficacy while maintaining adverse events at acceptable level of severity and incidence.
References:
1] Bai JPF et al. CPT:PSP (2024) 13(12), 2102-2110
[2] Chen DS, Mellman I. Immunity (2013) 39(1), 1-10
[3] Li L, Gardner I, Dostalek M, Jamei M. AAPS J. (2014) 16(5),1097-109.
[4] Betts A et al. AAPS J. (2019) 21(4), 66
[5] Allen RJ, Rieger TR, Musante CJ. CPT:PSP (2016) 5(3),140-6
[6] U.S. Food and Drug Administration. BLA 761344 Multidisciplinary Review and Evaluation. Amgen Inc. (2024)
[7] Chiang AC et al. Clin Lung Cancer. (2020) 21(5), 455-463.e4
[8] George J et al. Nature (2015) 524(7563), 47-53
[9] Giffin MJ et al. Clin Cancer Res. (2021) 27(5),1526-1537
[10] Antonia SJ et al. Lancet Oncol. (2016) 17(7),883-895
Reference: PAGE 34 (2026) Abstr 12028 [www.page-meeting.org/?abstract=12028]
Poster: Drug/Disease Modelling - Oncology