ANDROMEDA by Prosilico and log D outperform human hepatocytes for the prediction of intrinsic hepatic metabolic clearance of carboxylic acids

IntroductionExtrahepatic metabolism/conjugation, deconjugation of their metabolites, and low and varying unbound fraction in plasma (fu), is characteristic for carboxylic drugs. Thus, it is comparably difficult to estimate their in vivo intrinsic hepatic metabolic clearance (CLint) and hepatic CL (CLH) and to predict their in vivo CLint, CLH and CL. One objective was to investigate the laboratory variability of fu and CLint for carboxylic acids. Another objective was to compare human hepatocytes, measured log D and the software ANDROMEDA with regards to prediction of human in vivo CLint of carboxylic acids. Materials and MethodsMeasured unbound hepatocyte CLint, non-renal CL (surrogate for CLH), non-renal CLint (surrogate for hepatic metabolic CLint), log D and fu data were taken from studies in the literature. ANDROMEDA (by Prosilico; version 1.0) prediction software was used for in silico predictions of CLint for carboxylic acids not used in the training set of its CLint-model. Results and DiscussionMean and maximum differences between highest and lowest reported in vivo CLint predicted from hepatocyte CLint were 210- and 1,476-fold (n=8), respectively. Corresponding estimates for in vitro fu were 19- and 50-fold, respectively. The data set with the apparently highest number of carboxylic acids contains 39 carboxylic acids with in vitro CLint and log D (both measured at the same laboratory), in vivo CLint and in vitro fu. 18 carboxylic acids were excluded as their in vitro CLint was below the limit of quantification. The correlation coefficient (R2) for log hepatocyte predicted in vivo CLint vs log in vivo CLint was 0.34. The corresponding R2 for log D vs log in vivo CLint was 0.40 (0.47 for 64 carboxylic acids). The Q2 (forward-looking R2) for in silico (ANDROMEDA) predicted and measured log in vivo CLint for 12 carboxylic acids was 0.86. The corresponding R2 for hepatocytes and log D were 0.67 and 0.66, respectively. ANDROMEDA produced a lower maximum prediction error compared to hepatocytes and also predicted the in vivo CLint for all carboxylic acids out of reach for the hepatocyte assay. ConclusionVery large interlaboratory variability was demonstrated for plasma protein binding and hepatocyte assays. Log D, and especially ANDROMEDA, outperformed the hepatocyte assay for the prediction of CLint of carboxylic acids in vivo in man.