Impact of genotype assumption in a semi-mechanistic PK model of metformin
Gustaf J Wellhagen (1), Tore Bjerregaard Stage (1,2,3), Kim Brøsen (2), Maria C Kjellsson (1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Sweden, (2) Clinical pharmacology, Department of Public Health, University of Southern Denmark, (3) University of California San Francisco, San Francisco, USA
Objectives: Develop a semi-mechanistic model of metformin PK and investigate the impact of different ways of handling missing genetic variants in transporters on metformin pharmacokinetics.
Methods: A semi-mechanistic PK model of metformin was developed using data from three studies; one single- and two steady-state studies. Plasma concentration and urine samples were collected in 87 healthy volunteers. Subjects were genotyped for SNPs previously associated with metformin PK (MATE1 variants-rs2289669, rs2252281; MATE2-K variant-rs12943590; OCT1 variants-rs622342, reduced function variants (rs12208357, rs34130495, rs72552763 and rs34059508, O1R); OCT2 variant-rs316019; ATM-rs11212617) [1]. Information on two SNPs (MATE2-K and ATM) were missing in 49% of the individuals (84% of individuals in single-dose study) and missing genotypes were handled by (i) excluding incomplete cases, (ii) assigning to wildtype (wt), (iii) heterozygote variant (wt/v), (iv) homozygote variant (v/v) or (v) model-based estimation of genotype [2]. For MATE2-K , the background frequencies of variants were 55%, 35% and 10% of wt, wt/v and v/v, respectively and for ATM, the corresponding frequencies were 26.5%, 54% and 19.5%, which were used in approach (v).
Results: A 2-compartment model including renal clearance with saturable reabsorption, filtration proportional to GFR and active secretion (CL) best fitted the data. No covariates influenced absorption constant (KA), but CL increased by 0.7% per kg. Estimated bioavailability (F) was 49% for subjects with wildtype of O1R and 43% for subjects with the SNP. MATE1 variant-rs2289669, OCT1 variant-rs622342 and AGE was found to affect CL, however not when excluding incomplete cases. These covariates mainly improve the fit of single dose-PK. Depending on approach, different genotypes affected inter-compartmental clearance (Q): OCT2 variant when missing assumed to be wt, MATE2-K variant when missing assumed to be wt/v, ATM variant when missing assumed to be v/v or inferred by the model.
Conclusions: As expected, how the missing genotypes were handled influenced covariate inclusion. However, approach (i) also affected inclusion of other covariates as there was a correlation between missingness and study design. In this particular case, exclusion of incomplete cases removed data from the single-dose study which is less interesting for treatment of a chronic disease and the results from excluding incomplete cases was the most robust method.
References:
[1] Gong, Li et al. “Metformin Pathways: Pharmacokinetics and Pharmacodynamics.” Pharmacogenetics and genomics 22.11 (2012): 820–827.
[2] Johansson, Åsa M., and Mats O. Karlsson. “Comparison of Methods for Handling Missing Covariate Data.” The AAPS Journal 15.4 (2013): 1232–1241.
