Towards more Realistic Clinical Trial Simulation: Establishing Inter-Correlations between Several Cytochrome P450 Enzyme Abundances in Human Liver

Objectives: Success of clinical trial simulations greatly depends on identifying and incorporating true anatomical and physiological covariates when generating virtual populations. Recently, the inter-correlations between CYP450 (and UGT) metabolising enzymes in human liver microsomes have been reported [1, 2]. These data enable population-based PBPK models to more realistically assign CYP/UGT abundances when generating virtual subjects using Correlated Monte Carlo sampling. This work aims at using reported data in [1, 2] to determine CYP450 enzymes covariance matrix of multivariate probability distribution function and compare the distributions against the currently generated data in Simcyp Simulator version 14. Methods: The marginal distributions of each CYP450 are assumed to be log-normal and are checked by obtaining normal plots for the log transformed abundance value. The correlation and covariance matrices for the whole dataset have been calculated using R Package (v 3.1.2). Eigenvalues of the covariance matrix have been tested for positivity and if negative eigenvalues occurred, then the nearest positive definite covariance matrix has been calculated. A new correlated sample has been obtained by calculating the Cholesky decomposition [3] of the nearest positive definite covariance matrix and this sample has been transformed into its original log-normal marginal distribution. A virtual population matching the real population demographics has been generated using the Simcyp Simulator v 14 and the results were compared with Monte Carlo sampling data. Results: Multivariate Cramer-Test shows dissimilarity between the empirical distributions of the correlated and uncorrelated samples with 95% CI critical T-statistic of 141.565 vs observed: 9269.496. Kernel density estimate visually showed the differences between the two distributions. Conclusions: Incorporation of the correlated CYP450 enzyme abundances when generating virtual subjects enables PBPK simulators to generate more realistic virtual population which can improve clinical study design and prediction of clinical outcome during drug development. Further research is needed to establish various transporters abundances inter-correlations which can have a significant impact on the drug concentration at the site of action and as a result on drug safety and efficacy. References: [1] Brahim Achour, Matthew R. Russel, Jill Barber, and Amin Rostami-Hodjegan, Simultaneous Quantification of the Abundance of Several Cytochrome P450 and Uridine 59-Diphospho-Glucuronosyltransferase Enzymes in Human Liver Microsomes Using Multiplexed Targeted Proteomics, Drug Metab Dispos 42:500–510 (2014) [2] Brahim Achour, Jill Barber, and Amin Rostami-Hodjegan, Expression of Hepatic Drug-Metabolizing Cytochrome P450 Enzymes and Their Intercorrelations: A Meta-Analysis, Drug Metab Dispos 42:1349–1356 (2014) [3] Günther Hämmerlin, Karl-Heinz Hoffmann, Numerical Mathematics, Springer-Verlag New York Inc. (1991)