Adrien Tessier (1), Karl Brendel (1), Charlotte Gesson (1), Sylvain Fouliard (1), Emmanuelle Comets (2,3) and Marylore Chenel (1)
(1) Division of Clinical Pharmacokinetics and Pharmacometrics, Institut de Recherches Internationales Servier, Suresnes, France. (2) INSERM, IAME, UMR 1137 ; Université Paris Diderot, Sorbonne Paris Cité, F-75018 Paris, France. (3) INSERM CIC 1414, Université Rennes 1, Rennes, France.
Objectives: More and more genetic data are collected in clinical trials. Based on an example, we propose an approach to integrate a large number of genetic variants in population PK models, using a penalised regression approach recently evaluated in this context [1,2].
Methods: Drug S is a compound metabolised mainly by CYP1A2 (80%). This metabolism enzyme is highly polymorphic and numerous environmental factors (tobacco, estrogens…) can modify its expression [3], in addition to genetic factors. Data came from 10 phase I studies, including 257 healthy volunteers receiving single or repeated oral administrations. Several demographic (age, gender, BMI, BSA, ethnicity) and environmental (food, tobacco or contraceptives use) variables were measured. Subjects were also genotyped for 176 Single Nucleotide Polymorphisms (SNPs) using a DNA microarray designed for ADME genes. A population PK model was built on these data.
In the covariates model building, demographic and environmental covariates were first included in the PK model, so that genetic variants were next associated with the remaining unexplained variance of clearance (CL).
Then two pharmacogenetic (PG) analyses were performed: a targeted analysis, where only associations between 3 SNPs from CYP1A2 and CL were investigated through LRT [4]; and a blinded analysis, where associations between CL and the 176 SNPs were explored using a Lasso regression [5]. All genetic covariates selected by the Lasso were included and tested in the PK model through LRT.
Estimation of the population parameters was performed using NONMEM 7.3 and FOCE-I algorithm. Lasso analysis was performed using the HyperLasso program [6].
Results: Drug S PK data were described by a two-compartment model, with a linear elimination. Two shifted first-order absorptions were used to describe a late rebound in concentrations and bioavailability parameter F was nonlinear with dose.
Effects of food, gender, age, ethnicity and contraceptives intake were included in the PK model. One third of the CL variance was explained by these covariates. None of the two PG analyses showed significant association between genetic variants and the remaining unexplained CL variance.
Conclusion: In this approach, genetic variants, which are often associated with low to moderate effects on PK, were explored taking into account other confounding covariates. This approach could allow detecting smaller genetic effects. For drug S, no polymorphism was associated with the unexplained interindividual variability.
References:
[1] Bertrand et al. Pharmacogenet Genomics. 2013
[2] Tessier et al. AAPS J. 2015
[3] Faber et al. Basic Clin Pharmacol Toxicol. 2005
[4] Lehr et al. Pharmacogenet Genomics. 2010
[5] Tibshirani. J R Stat Soc Ser B. 1994
[6] Hoggart. PLoS Genet. 2008
Reference: PAGE 25 () Abstr 5948 [www.page-meeting.org/?abstract=5948]
Poster: Drug/Disease modeling - Other topics