Arthur Van De Vyver (1), Miro Eigenmann (1), Sylvia Herter (2),Jitka Somandin (2), Nicolas Frances (1), Marina Bacac (2), Antje-Christine Walz (1)
Roche Pharma Research and Early Development, (1) Pharmaceutical Sciences, Roche Innovation Center Basel; (2) Cancer Immunotherapy department 2, Roche Innovation Center Zürich
Introduction: T-Cell Bispecifics (TCBs) are antibody constructs with binding specificities to both a tumor antigen and T cell receptor allowing for cross-linking and resulting in the formation of immunological synapses between T- and tumor cells, and subsequent T cell mediated tumor cell lysis. Existing mathematical models link the trimolecular binding of drug with tumor antigen and with CD3-receptor on T cells to tumor cell killing and are aimed at guiding compound optimization [1], providing a new MABEL metric (Minimal Anticipated Biological Effect Level) for dose selection [2], or describing in vivo tumor cell and T cell profiles [3]. Here, we compare model structures that capture relevant processes to predict tumor cell killing with Cibisatamab, a novel TCB targeting carcinoembryonic antigen (CEA) in high and low tumor target expressing cells [4]. We therefore fitted various models to an in vitro PKPD study capturing longitudinal data from tumor cell and T-cell dynamics upon Cibisatamab treatment in 2 different cells lines with high and low CEA expression exhibiting distinct efficacy profiles. While full tumor cell killing, T-cell activation and expansion was observed in the high expressing cell line, only partial killing was observed in low expressing cell line with minimal T-cell activation but no T cell expansion.
Objective: The goal of this work was to compare various model variants in their capacity to predict how TCB drives tumor cell killing in function of TCB concentration and target expression levels, observed in vitro with and without consideration of immunological synapse formation and T cell dynamics.
Methods: Three model types with increasing complexity in terms of number of differential equations and parameters were compared. Model 1 is a simple model, directly linking TCB concentration to tumor cell killing [5]. Model 2 & 3 are model variants based on a TCB synapse model from Jiang et al.[1]. These models assume immunological synapse formation to take place due to interaction of TCB with tumor antigen and T cell CD3 as independent binding events. In the first synapse model (model 2), synapse formation will drive tumor cell killing. In the second synapse model (model 3), synapse formation will lead to T cell activation and expansion, which will drive tumor cell killing.
These models were fitted to full time course data from an in vitro study where 2 tumor cell lines with different CEA expression levels, MKN45 (230’000-690’000 CEA/cell) and Cx1 (2’000-11’000 CEA/cell), were co-cultured with hPBMCs (human peripheral blood mononuclear cells) at different Cibisatamab concentrations ranging from 6 to 100’000 pM[4][6]. Tumor cell and T cell counts were measured 24h, 48h, 72h, 96h, and 168h after adding Cibisatamab to the co-culture. Total number of observations was 160, tested in 2 cell lines at 8 dose subgroups monitored at 5 time points. Model fitting was performed in Monolix (2018R1).
The quality of model fitting was assessed by visual inspection and goodness of fit criteria such as observed versus predicted values, visual predictive checks of observations and predictions at each dose, the precision of parameter estimation, and the reduction in objective function values.
Results:
Best model fits were obtained with model 3 (synapse model with T cell dynamics), based on a reduction in objective function values and visual inspection of the fits (observed vs. predicted values, prediction distribution). The simpler model linking drug concentration to tumor killing was not sufficient to accurately describe the tumor kill profiles, and was unable to correctly describe the killing of the low CEA expressing tumor. Both models including synapse formation described tumor killing reasonably well. In addition, the model considering T cell activation (model 3) showed enhanced model performance as indicated by improved objective function values. This is supported by significantly narrowed prediction intervals, especially at higher TCB concentrations.
Conclusion: In the present study, the consideration of synapse formation significantly improved prediction of anti-tumor effects. Considering T cell profiles as part of the mechanistic model further improved performance. Furthermore, it has the potential to allow us to predict tumor cell killing under various T cell conditions in vitro, this can enable us to predict the outcome of varying Effector-to-Target ratios and improve translation to in vivo.
References:
[1] Jiang, X., et al., Development of a Target cell-Biologics-Effector cell (TBE) complex-based cell killing model to characterize target cell depletion by T cell redirecting bispecific agents. MAbs, 2018. 10(6): p. 876-889.
[2] Chen, X., et al., Mechanistic Projection of First-in-Human Dose for Bispecific Immunomodulatory P-Cadherin LP-DART: An Integrated PK/PD Modeling Approach. Clin Pharmacol Ther, 2016. 100(3): p. 232-41.
[3] Campagne, O., et al., Integrated Pharmacokinetic/Pharmacodynamic Model of a Bispecific CD3xCD123 DART Molecule in Nonhuman Primates: Evaluation of Activity and Impact of Immunogenicity. Clin Cancer Res, 2018. 24(11): p. 2631-2641.
[4] Bacac, M., et al., A Novel Carcinoembryonic Antigen T-Cell Bispecific Antibody (CEA TCB) for the Treatment of Solid Tumors. Clin Cancer Res, 2016. 22(13): p. 3286-97.
[5] Lobo, E.D. and J.P. Balthasar, Pharmacodynamic modeling of chemotherapeutic effects: application of a transit compartment model to characterize methotrexate effects in vitro. AAPS pharmSci, 2002. 4(4): p. E42-E42.
[6] Eigenmann M et al. PAGE 27 (2018) Abstr 8621 [www.page-meeting.org/?abstract=8621]
Reference: PAGE 28 (2019) Abstr 9077 [www.page-meeting.org/?abstract=9077]
Poster: Drug/Disease Modelling - Oncology