(W0930-11-61) Setting up Success in the Clinic through Semi-mechanistic Translational Modelling

Purpose: During the development of Oncology compounds the main aim of clinical dose finding studies was to find the maximum tolerable dose (MTD). Recently, FDA’s Oncology Centre of Excellence launched Project Optimus[1], a new guidance on dose optimisation for Oncology therapeutics. The guidance signals a cultural shift [2], in Oncology drug development, from finding a tolerable limit to optimizing a dose which is efficacious and safe in patients. Translational modelling approaches, integrated with experimental data, serve as a powerful real-time tool to support dose optimisation and justification. There remains challenges and uncertainties in translating preclinical exposure-response relationships to clinical response. In this poster we demonstrate a novel approach to overcoming translational challenges through a semi-mechanistic model linking pharmacokinetics (PK), pharmacodynamics (PD) and tumour response (TGI) i.e. a PK-PD-TGI model.

Methods: We demonstrated a semi-mechanistic translational modelling approach on Sotorasib (AMG510) literature data [3,4] and qualified the modelling approach on Adagrasib (MRTX-849) literature data [5]. The first key step was to establish a preclinical PK-PD-TGI model for Sotorasib on data reported by Canon et al and Lanman et al. The PK-PD-TGI model was established in a stepwise manner: regress a PK model on the PK data, model pERK (a downstream biomarker of KRAS G12C) modulation in MIA-PaCa-2 cell-line to establish the PK-PD relationship, and finally model efficacy in MIA-PaCa-2 xenograft mice to estimate the tumour growth kinetics. Once, the structure of the model has been established preclinically it can then be translated to allow for clinical trial response prediction and optimization of dose for patients. Several sources of data for drug efficacy, such as in vitro potency or in vivo biomarker modulation and preclinical or clinical tumour growth kinetics, were implemented in the process of model creation. The final selected data sources were for those that gave the best prediction of Sotorasib efficacy. The mouse PK model is swapped out for a human PK model [6] and plasma protein binding accounted for. The IC50 value estimated on pERK modulation, in G12C sensitive MIA-PaCa-2, data is swapped for an IC50 distribution, derived from in vitro cell proliferation assays, to allow for a distribution prediction across a patient population. Tumour growth kinetics were derived from by back-calculating growth and decay rates based upon reported clinical response data [7]. Once the translational model had been established, we predicted the response rate, in accordance with the RECIST criteria, of a Phase II clinical trial of Sotorasib to assess the models predictive capability. We then qualified the model on Adagrasib clinical response data by swapping Sotorasib human PK model for Adagrasib human PK model [8] and deriving Adagrasib potency from a screening of in vitro cell proliferation data.

Results: A preclinical PK-PD-TGI model was successfully established on the literature data. A 2-compartment PK model with non-dose proportional exposure captured the mean PK profiles at steady state in a satisfactory manner. The exposure-response model was established on pERK modulation in the, G12C sensitive, MIA-PaCa-2 cell-line. Tumour growth inhibition in MIA-PacA-2 cell-line was well captured by the PK-PD-TGI model where pERK modulation drove tumour growth inhibition. After humanizing the PK-PD-TGI, by swapping mouse PK model for the human PK model the predicted response rate, 99.9% (99.2%, 100%) was significantly higher than the observed response rate, 37.1% (28.6%, 46.2%) in the Phase II trial of non-small cell lung cancer (NSCLC) patients. This was to be expected as the exposure-response rate was estimated on highly sensitive cell-line as opposed to a heterogenous patient population. The model was iterated, by deriving an IC50 distribution from in vitro cell proliferation data and clinical tumour growth kinetics from Sotorasib tumour response data. This led to a greatly improved and satisfactory prediction of the clinical tumour response in G12C mutated NSCLC patients, 42.6% (34.7%, 51.6%). Adagrasib human PK model and IC50 distribution, derived off in vitro cell proliferation data, was integrated into the PK-PD-TGI model and a response rate of 41.2% (20.0%, 66.7%) which captures the observed clinical response, 53.3% (26.6%, 78.7%), thus a translational model has been successfully established and qualified.

Conclusion: We have proposed a method to translate a preclinical semi-mechanistic PK-PD-TGI model, where target modulation drives tumour growth inhibition, to a clinical PK-PD-TGI model that can be used to predict the response rate of a given dose to a patient at the first assessment in the clinic. We successfully applied the framework to Sotorasib data and qualified the model on, another KRAS 12C inhibitor, Adagrasib. This translational approach will allow for an improved of selection of lead compounds and the optimal clinical exposure/dose to be identified and justified thus increasing the chances of success and aiding in regulatory interactions.

References: https://www.fda.gov/about-fda/oncology-center-excellence/project-optimus
Optimize the dose: An optimal step forward for FDA, The Cancer Letter, Vol. 47 No.23, 2021
Canon et al 2019, DOI: https://doi.org/10.1038/s41586-019-1694-1
Lanman et al 2019, DOI: 10.1021/acs.jmedchem.9b01180
Hallin et al 2020, DOI: 10.1158/2159-8290
FDA’s multi-discipline review, application number 214665Orig1s000
Skoulidis et al 2021, DOI: 10.1056/NEJMoa2103695
Ou et al 2022, DOI: 10.1200/JCO.21.02752

Acknowledgements: The authors greatly appreciate comments, helpful suggestions from and valuable discussion with the following colleagues at Boehringer-Ingelheim: Dr. Jens Quant, Dr. Otmar Schaaf, Dr. Fabio Savarese, Dr. Jan Elias, Dr. Lu Vivian Tan, Mr. Matthias Klemencic, Dr. Johannes Popow and Dr. Karl-Heinz Liesenfeld


Figure 1 Translational modelling concept


Figure 2 Preclinical PK-PD-TGI modelling of Sotorasib literature data


Figure 3 Prediction of clinical response rate in NSCLC patients after administration of Sotorasib or Adagrasib