Zinnia P Parra-Guillen (1,2), Johan E Wallin (3), Celine Pitou (3), Philip W Iversen (4), Carmine Carpenito (4), David Surguladze (4), Ivan Inigo (4), Darin Chin (4), Iñaki F Troconiz (1,2)
(1) Pharmacometrics & Systems Pharmacology Group, Department of Pharmaceutical Technology and Chemistry, School of Pharmacy and Nutrition, University of Navarra, Pamplona, Spain, (2) IdiSNA, Navarra Institute for Health Research, Pamplona, Spain, (3) Global PK/PD & Pharmacometrics, Eli Lilly and Company, (4) Lilly Research Laboratories, Eli Lilly and Company
Objectives: Humanised mouse models are widely used for preclinical evaluation of immuno-oncological (IO) drugs [1]. However, tumour response from these models is highly variable and dependent on the donor properties, potentially masking interpretation of drug effects. The objective of this work was to propose a mathematical framework to quantitatively assess (i) the in vivo effect of different therapeutic strategies during early preclinical development and (ii) the variability associated to these experimental settings to support drug development.
Methods: Longitudinal tumour volume from 22 different studies performed in established HCC827 xenograft tumour model were available. Briefly, NSG mice were implanted subcutaneously with 107 HCC827 tumour cells. When tumours reached a predefined volume (~300mm3), a single intraperitoneal (IP) dose of human IgG (n=160) or a single intravenous dose of expanded human T cells (dose ranging between 2 to 4×106) was intravenously administered. After T cell administration, mice were left untreated (n=229) or were treated with weekly IP injections of different IO compounds under research for 4 weeks (n=769). In total, tumour growth data after administration of T cells from 9 different donors, and 6 different IO drugs was collected. In addition, pharmacokinetic data from all drugs was available. A stepwise population modelling approach was followed using NONMEM 7.3 and FOCEI algorithm: (i) tumour growth modelling in the absence of perturbation (Tcells or drug administration) was first characterised, (ii) T cell effect modelling was developed and finally (iii) treatment effect modelling was implemented. Additional levels of variability above inter-animal variability (IAV), i.e. inter-study variability (ISV) and/or inter-donor variability (IDV), were explored.
Results: The unperturbed tumour growth model proposed by Simeoni et al [2] enabled a satisfactory characterization of tumour data in absence of treatment, with an adequate parameter precision (relative standard errors, RSE, below 10%) and low inter-animal variability (IAV, below 25%). In addition, low inter-study variability at initial tumour burden was detected (ca. 20%) explaining half of the IAV at this level. Regarding the T cell model, a simeoni-like structure -where administered T cells trigger tumour cell death through a series of transit compartments- provided a better overall characterisation over alternative models, and was thus selected as final T cell model. Due to the lack of T cells measurements, T cells were assumed to engraft and remain constant over time at the administered dose. An efficacy parameter (KTcell) value of 0.013 (106 cells*day)-1 with a relatively high associated IAV of 105% was estimated with high precision (RSE <10%). A relatively large inter-study variability (50 %) was identified explaining around 20 % of the variability estimated on KTcell. Unfortunately, using donor information instead of study information did not provide a better model performance or variability explanation, probably due to the lower number of donors compared to the number of studies and the high intrinsic IAV. Finally, and in agreement with the known biology, drug effects were introduced implemented in the model increasing T cells activity (i.e. no drug effect in the absence of T cell administration). For all drug candidates except one, a non-linear drug effect model (EMAX model) provided a better description than the linear relationship. Additional IAV could not be identified at this step. Overall, a good characterisation of all treatment scenarios was achieved using the common T cell model structured and estimating drug-specific parameters (EMAX and EC50).
Conclusions: A quantitative framework has been developed to describe the effects of different onco-immunological treatments during early preclinical development using a common model structure. This methodology enables the ranking and comparison of different IO drug candidates, together with the quantification of the different sources of variability, thus facilitating data interpretation. Moreover, the identification of drug-specific parameters opens the possibility to explore potential in vitro-in vivo correlations that could be used to predict in vivo drug efficacy from in vitro experiments, as well as to extrapolate preclinical results to human scenarios.
References:
[1] Zitvoge et al. Nat Rev Cancer 2016
[2] Simeoni et al. Cancer Research 2004
Reference: PAGE 28 (2019) Abstr 9156 [www.page-meeting.org/?abstract=9156]
Poster: Drug/Disease Modelling - Oncology