Computational analysis of cytokine release following bispecific T-cell engager therapy
Gianluca Selvaggio (1), Silvia Parolo (1), Pranami Bora (1), Lorena Leonardelli (1), John Harrold (5,3), Dan Rock (5,4), Khamir Mehta (2), Luca Marchetti (1)
(1) Fondazion The Microsoft Research - University of Trento Centre for Computational and Systems Biology (COSBI), Rovereto, TN – Italy; (2) Clinical Pharmacology, Modeling and Simulation, Amgen Inc, South San Francisco, CA – USA; (3) Current affiliation: Seattle Genetics, San Francisco, CA – USA; (4) Current affiliation: Merck, San Francisco, CA – USA, 5) Previous affiliation: Drug Metabolism and Pharmacokinetics, Amgen Inc, South San Francisco, CA – USA
Introduction: Bispecific T cell Engager (BiTE®) molecules offer a promising treatment that uses the patient’s own immune system towards eliminating cancerous cells. Realizing the full potential of this therapy requires the mitigation of the adverse effects of cytokine release from the immune activation that can eventually lead to cytokine release syndrome (CRS) [1].
Objectives:
- Develop a logic based computational model describing the interplay between BiTE® molecules, the immune system, and the tumor cells.
- Perform in silico simulations to improve the understanding of the factors affecting the magnitude of cytokine release.
- Explore possible co-treatments to administer with BiTE® molecules, that can help alleviate the cytokine release and minimize impact on efficacy.
Methods: The logical model was built using GINsim [2], a tool dedicated to the logical formalism, focusing on the major variables influencing the activation of the cytokine release and the tumor cell elimination. The rules governing the interactions were defined based on known biology and published experimental evidence. Each variable was discretized and was assigned minimal number of levels necessary to adequately represent its behavior. The severity of the inflammation was described by the number of cytokines at their maximum value (Combined Cytokine Index, CCI), used as proxy for the CRS observed in the clinical setting. The model stable states were assessed using GINsim built in functions, while a more quantitative description of the dynamics was obtained using the software MaBoSS [3]. The efficacy of co-treatments was assessed performing a sensitivity analysis of CCI and Tumor with respect to model mutation.
Results: The asymptotic behavior of the model shows that BiTE® molecules can be effective in eradicating the tumor. The model predicts differences in the dynamic behavior of the system, showing a dose effect for tumor clearance time and CCI intensity. In particular, the model predicts that higher the dose the faster the tumor clearance but also the stronger the CCI response. To assess how to ameliorate the inflammatory response without impairing the tumor killing capacity, we performed a systematic sensitivity analysis by mutating the model components (i.e., over-expression/knock-down). The model suggests IFN-γ as an optimal target, with either partial or total inhibition providing a significant benefit with respect to the wild type model. The sensitivities, also, highlighted that targeting TNFα or IL6 influences CCI, in agreement with the observations in literature [4]. We also performed a mutation of the rates of activation and deactivation of each node. The analysis identified IFN-γ as optimal target, highlighting that inhibition of the production rate would have a stronger effect than increasing its clearance. We also tested the effect of an early or late administration of BiTE® molecules with anti-TNFα or anti-PDL1. The latter, if administered after BiTE®, showed an improvement of the CCI, and a slower tumor clearance. On the contrary, anti-TNFα had the higher effect on CCI the sooner it was administered.
Conclusions: The current work showed how logical QSP models can be used to investigate complex phenomena and generate new testable hypotheses. Computational approaches can be instrumental to systematically analyze the effects of combining BiTE® molecules and CRS mitigation strategies on the tumor killing efficacy and cytokine release. Our analysis suggests that IFN-γ may be a good mechanism to control CRS risk in patients. Furthermore, it entails the existence of a time window to administrate anti-PDL1 therapy and mitigate CCI without compromising tumor clearance.
References:
[1] Liu, D. & Zhao, J. Cytokine release syndrome: Grading, modeling, and new therapy. J. Hematol. Oncol. 11, (2018).
[2] Naldi, A. et al. Logical Modeling and Analysis of Cellular Regulatory Networks With GINsim 3.0. Front. Physiol. 9, 646 (2018).
[3] Stoll, G., Viara, E., Barillot, E. & Calzone, L. Continuous time boolean modeling for biological signaling: application of Gillespie algorithm. BMC Syst. Biol. 6, 116 (2012).
[4] Li, J. et al. CD3 bispecific antibody-induced cytokine release is dispensable for cytotoxic T cell activity. Sci. Transl. Med. 11, (2019).