Using mixed effects modelling improves detection of drug-gene interactions in mouse trials
Emma C. Martin (1), Leon Aarons (1), Hitesh Mistry (1), James W.T. Yates (2)
(1) Centre for Applied Pharmacokinetic Research, Manchester Pharmacy School, the University of Manchester, M13 9PT, U.K (2) AstraZeneca, Innovative Medicines, Oncology, Modelling and Simulation, Hodgkin Building, Chesterford Research Park, Cambridge, CB10 1XL, U.K
Objectives: Mouse trials use multiple patient derived xenografts (PDXs) in order to reflect the heterogeneity in clinical cancer trials and to allow identification of subgroups of patients for treatment. Each PDX is treated once with each of the drugs being investigated, in a 1x1x1 design. These studies allow for the detection of drug-gene interactions, where gene mutations have an effect on drug-sensitivity. We aim to identify the best metric for measuring drug-sensitivity in order to identify drug-gene interactions.
Methods: The breast cancer PDXs from the data published by Gao et al. [1] was used to develop and test the methods. Three metrics considered were final tumour volume [2], mRECIST criteria [1] and growth rate in an exponential growth model. A mixed effect modelling approach with the interaction estimated as a parameter was also tested. The estimates of the interactions using each of the methods were compared to those found in a systematic literature review.
Results: Using a mixed effects modelling approach gave the most power to detect interactions. Both the modelling and growth rate approaches could be used to evaluate all interactions, even when data for a given animal was sparse, whereas missing data and drop outs led to a large number of interactions being non-evaluable for the other two methods. When compared to the 16 interactions found in the literature, 14 were correctly identified by the modelling and growth rate methods, the other two methods had a lower success rate, with fewer interactions available to compare.
Conclusions: Using modelling can help to identify interactions when using a mouse trial design. Changes to the design could potentially further aid in the identification of interactions. These include more replicates per patient tumour, particularly in the control group, and the inclusion of more dose levels for each treatment instead of a single clinically relevant or maximum tolerated dose per drug. Comparing the results to previously reported interactions in the literature allows for investigation into whether the results of mouse trials could be used to predict interactions which will later be confirmed clinically or preclinically, however there are potential issues with reproducibility [3] and publication bias [4].
References:
[1] Gao, H., et al., High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response. Nat Med, 2015. 21(11): p. 1318-25
[2] Zhang, X.C., et al., Establishment of patient-derived non-small cell lung cancer xenograft models with genetic aberrations within EGFR, KRAS and FGFR1: useful tools for preclinical studies of targeted therapies. J Transl Med, 2013. 11: p. 168.
[3] Baker, M. and E. Dolgin, Cancer reproducibility project releases first results. Nature, 2017. 541(7637): p. 269-270.
[4] Fanelli, D., Negative results are disappearing from most disciplines and countries. Scientometrics, 2012. 90(3): p. 891-904.