Fabesetron

Molecular modeling approaches for the prediction of the nonspecific binding of drugs to hepatic microsomes were examined using a published database of 56 compounds. Models generated were evaluated using an independent test set of 13 compounds. A pharmacophore approach identified structural features of drugs associated with nonspecific binding. A side-chain amino group and complementary hydrophobic domain were the principal features noted. The use of shape overlays, based on the pharmacophore, in conjunction with a chemical force field in the program ROCS, yielded discrimination between molecules classified as strong binders (experimental fraction unbound in microsomes<0.50) and those with a lower degree of binding (experimental fraction unbound in microsomes>0.50). In the initial data set of 56 molecules, 18 were classified as strong binders (on the basis of the above criteria), and all of those were recovered in the top 22 molecular hits from ROCS. Additionally, computationally generated values of log P were shown to provide a reasonable estimate of the fraction unbound in microsomes, providing the compounds were in their basic form at physiological pH.

The aim of this study was to evaluate a unified method for predicting human in vivo intrinsic clearance (CL(int, in vivo)) and hepatic clearance (CL(h)) from in vitro data in hepatocytes and microsomes by applying the unbound fraction in blood (fu(b)) and in vitro incubations (fu(inc)). Human CL(int, in vivo) was projected using in vitro data together with biological scaling factors and compared with the unbound intrinsic clearance (CL(int, ub, in vivo)) estimated from clinical data using liver models with and without the various fu terms. For incubations conducted with fetal calf serum (n=14), the observed CL(int, in vivo) was modeled well assuming fu(inc) and fu(b) were equivalent. CL(int, ub, in vivo) was predicted best using both fu(b) and fu(inc) for other hepatocyte data (n=56; r(2)=0.78, p=3.3 x 10(-19), average fold error=5.2). A similar model for CL(int, ub, in vivo) was established for microsomal data (n=37; r(2)=0.77, p=1.2 x 10(-12), average fold error=6.1). Using the model for CL(int, ub, in vivo) (including a further empirical scaling factor), the CL(h) in humans was also calculated according to the well stirred liver model for the most extensive dataset. CL(int, in vivo) and CL(h) were both predicted well using in vitro human data from several laboratories for acidic, basic, and neutral drugs. The direct use of this model using only in vitro human data to predict the metabolic component of CL(h) is attractive, as it does not require extra information from preclinical studies in animals.